cmd/tailscale: add basic support for admin subcommand

The admin subcommand is a thin wrapper over the REST API.
It (hopefully) makes administration of tailnets easier
than vanilla curl.

Signed-off-by: Joe Tsai <joetsai@digital-static.net>

.github/ISSUE_TEMPLATE/bug_report.md

@@ -2,36 +2,7 @@
 name: Bug report
 about: Create a bug report
 title: ''
-labels: ''
+labels: 'needs-triage'
 assignees: ''
 
 ---
-
-<!-- Please note, this template is for definite bugs, not requests for
-support. If you need help with Tailscale, please email
-support@tailscale.com. We don't provide support via Github issues. -->
-
-**Describe the bug**
-A clear and concise description of what the bug is.
-
-**To Reproduce**
-Steps to reproduce the behavior:
-1. Go to '...'
-2. Click on '....'
-3. Scroll down to '....'
-4. See error
-
-**Expected behavior**
-A clear and concise description of what you expected to happen.
-
-**Screenshots**
-If applicable, add screenshots to help explain your problem.
-
-**Version information:**
- - Device: [e.g. iPhone X, laptop]
- - OS: [e.g. Windows, MacOS]
- - OS version: [e.g. Windows 10, Ubuntu 18.04]
- - Tailscale version: [e.g. 0.95-0]
-
-**Additional context**
-Add any other context about the problem here.

.github/ISSUE_TEMPLATE/feature_request.md

@@ -2,25 +2,6 @@
 name: Feature request
 about: Suggest an idea for this project
 title: ''
-labels: ''
+labels: 'needs-triage'
 assignees: ''
-
 ---
-
-**Is your feature request related to a problem? Please describe.**
-
-A clear and concise description of what the problem is. Ex. I'm always
-frustrated when [...]
-
-**Describe the solution you'd like**
-
-A clear and concise description of what you want to happen.
-
-**Describe alternatives you've considered**
-
-A clear and concise description of any alternative solutions or
-features you've considered.
-
-**Additional context**
-
-Add any other context or screenshots about the feature request here.

.github/workflows/coverage.yml

@@ -1,48 +0,0 @@
-name: Code Coverage
-
-on:
-  push:
-    branches:
-      - main
-  pull_request:
-    branches:
-      - '*'
-
-jobs:
-  build:
-    runs-on: ubuntu-latest
-
-    if: "!contains(github.event.head_commit.message, '[ci skip]')"
-
-    steps:
-
-    - name: Set up Go
-      uses: actions/setup-go@v1
-      with:
-        go-version: 1.16
-      id: go
-
-    - name: Check out code into the Go module directory
-      uses: actions/checkout@v1
-
-      # https://markphelps.me/2019/11/speed-up-your-go-builds-with-actions-cache/
-    - name: Restore Cache
-      uses: actions/cache@preview
-      id: cache
-      with:
-        path: ~/go/pkg/mod
-        key: ${{ runner.os }}-${{ hashFiles('**/go.sum') }}
-
-    - name: Basic build
-      run: go build ./cmd/...
-
-    - name: Run tests on linux with coverage data
-      run: go test -race -coverprofile=coverage.txt -bench=. -benchtime=1x ./...
-
-    - name: coveralls.io
-      uses: shogo82148/actions-goveralls@v1
-      env:
-        COVERALLS_TOKEN: ${{ secrets.COVERALLS_TOKEN }}
-        GITHUB_TOKEN: ${{ secrets.COVERALLS_BOT_PUBLIC_REPO_TOKEN }}
-      with:
-        path-to-profile: ./coverage.txt

.github/workflows/go_generate.yml

@@ -0,0 +1,34 @@
+name: go generate
+
+on:
+  push:
+    branches:
+      - main
+      - "release-branch/*"
+  pull_request:
+    branches:
+      - "*"
+
+jobs:
+  check:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Set up Go
+        uses: actions/setup-go@v1
+        with:
+          go-version: 1.16
+
+      - name: Check out code
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 0
+
+      - name: check 'go generate' is clean
+        run: |
+          mkdir gentools
+          go build -o gentools/stringer golang.org/x/tools/cmd/stringer
+          PATH="$PATH:$(pwd)/gentools" go generate ./...
+          echo
+          echo
+          git diff --name-only --exit-code || (echo "The files above need updating. Please run 'go generate'."; exit 1)

.github/workflows/linux-race.yml

@@ -0,0 +1,48 @@
+name: Linux race
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - '*'
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    if: "!contains(github.event.head_commit.message, '[ci skip]')"
+
+    steps:
+
+    - name: Set up Go
+      uses: actions/setup-go@v1
+      with:
+        go-version: 1.16
+      id: go
+
+    - name: Check out code into the Go module directory
+      uses: actions/checkout@v1
+
+    - name: Basic build
+      run: go build ./cmd/...
+
+    - name: Run tests and benchmarks with -race flag on linux
+      run: go test -race -bench=. -benchtime=1x ./...
+
+    - uses: k0kubun/action-slack@v2.0.0
+      with:
+        payload: |
+          {
+            "attachments": [{
+              "text": "${{ job.status }}: ${{ github.workflow }} <https://github.com/${{ github.repository }}/commit/${{ github.sha }}/checks|${{ env.COMMIT_DATE }} #${{ env.COMMIT_NUMBER_OF_DAY }}> " +
+                      "(<https://github.com/${{ github.repository }}/commit/${{ github.sha }}|" + "${{ github.sha }}".substring(0, 10) + ">) " +
+                      "of ${{ github.repository }}@" + "${{ github.ref }}".split('/').reverse()[0] + " by ${{ github.event.head_commit.committer.name }}",
+              "color": "danger"
+            }]
+          }
+      env:
+        SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}
+      if: failure() && github.event_name == 'push'
+

.github/workflows/linux.yml

@@ -29,7 +29,7 @@ jobs:
       run: go build ./cmd/...
 
     - name: Run tests on linux
-      run: go test ./...
+      run: go test -bench=. -benchtime=1x ./...
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/linux32.yml

@@ -29,7 +29,7 @@ jobs:
       run: GOARCH=386 go build ./cmd/...
 
     - name: Run tests on linux
-      run: GOARCH=386 go test ./...
+      run: GOARCH=386 go test -bench=. -benchtime=1x ./...
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/staticcheck.yml

@@ -24,11 +24,23 @@ jobs:
     - name: Run go vet
       run: go vet ./...
 
-    - name: Print staticcheck version
-      run: go run honnef.co/go/tools/cmd/staticcheck -version
+    - name: Install staticcheck
+      run: "GOBIN=~/.local/bin go install honnef.co/go/tools/cmd/staticcheck"
 
-    - name: Run staticcheck
-      run: "go run honnef.co/go/tools/cmd/staticcheck -- $(go list ./... | grep -v tempfork)"
+    - name: Print staticcheck version
+      run: "staticcheck -version"
+
+    - name: Run staticcheck (linux/amd64)
+      run: "GOOS=linux GOARCH=amd64 staticcheck -- $(go list ./... | grep -v tempfork)"
+
+    - name: Run staticcheck (darwin/amd64)
+      run: "GOOS=darwin GOARCH=amd64 staticcheck -- $(go list ./... | grep -v tempfork)"
+
+    - name: Run staticcheck (windows/amd64)
+      run: "GOOS=windows GOARCH=amd64 staticcheck -- $(go list ./... | grep -v tempfork)"
+
+    - name: Run staticcheck (windows/386)
+      run: "GOOS=windows GOARCH=386 staticcheck -- $(go list ./... | grep -v tempfork)"
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/windows-race.yml

@@ -0,0 +1,55 @@
+name: Windows race
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - '*'
+
+jobs:
+  test:
+    runs-on: windows-latest
+
+    if: "!contains(github.event.head_commit.message, '[ci skip]')"
+
+    steps:
+
+    - name: Install Go
+      uses: actions/setup-go@v2
+      with:
+        go-version: 1.16.x
+
+    - name: Checkout code
+      uses: actions/checkout@v2
+
+    - name: Restore Cache
+      uses: actions/cache@v2
+      with:
+        path: ~/go/pkg/mod
+        key: ${{ runner.os }}-go-${{ hashFiles('**/go.sum') }}
+        restore-keys: |
+          ${{ runner.os }}-go-
+
+    - name: Test with -race flag
+      # Don't use -bench=. -benchtime=1x.
+      # Somewhere in the layers (powershell?)
+      # the equals signs cause great confusion.
+      run: go test -race -bench . -benchtime 1x ./...
+
+    - uses: k0kubun/action-slack@v2.0.0
+      with:
+        payload: |
+          {
+            "attachments": [{
+              "text": "${{ job.status }}: ${{ github.workflow }} <https://github.com/${{ github.repository }}/commit/${{ github.sha }}/checks|${{ env.COMMIT_DATE }} #${{ env.COMMIT_NUMBER_OF_DAY }}> " +
+                      "(<https://github.com/${{ github.repository }}/commit/${{ github.sha }}|" + "${{ github.sha }}".substring(0, 10) + ">) " +
+                      "of ${{ github.repository }}@" + "${{ github.ref }}".split('/').reverse()[0] + " by ${{ github.event.head_commit.committer.name }}",
+              "color": "danger"
+            }]
+          }
+      env:
+        SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}
+      if: failure() && github.event_name == 'push'
+

.github/workflows/windows.yml

@@ -33,7 +33,10 @@ jobs:
           ${{ runner.os }}-go-
 
     - name: Test
-      run: go test ./...
+      # Don't use -bench=. -benchtime=1x.
+      # Somewhere in the layers (powershell?)
+      # the equals signs cause great confusion.
+      run: go test -bench . -benchtime 1x ./...
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/xe-experimental-vm-test.yml

@@ -0,0 +1,45 @@
+name: "integration-vms"
+
+on:
+  pull_request:
+    paths:
+      - "tstest/integration/vms/**"
+  release:
+    types: [ created ]
+
+jobs:
+  experimental-linux-vm-test:
+    # To set up a new runner, see tstest/integration/vms/runner.nix
+    runs-on: [ self-hosted, linux, vm_integration_test ]
+
+    if: "!contains(github.event.head_commit.message, '[ci skip]')"
+
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v1
+
+      - name: Download VM Images
+        run: go test ./tstest/integration/vms -run-vm-tests -run=Download -timeout=60m -no-s3
+        env:
+          XDG_CACHE_HOME: "/var/lib/ghrunner/cache"
+
+      - name: Run VM tests
+        run: go test ./tstest/integration/vms -v -run-vm-tests
+        env:
+          TMPDIR: "/tmp"
+          XDG_CACHE_HOME: "/var/lib/ghrunner/cache"
+
+      - uses: k0kubun/action-slack@v2.0.0
+        with:
+          payload: |
+            {
+              "attachments": [{
+                "text": "${{ job.status }}: ${{ github.workflow }} <https://github.com/${{ github.repository }}/commit/${{ github.sha }}/checks|${{ env.COMMIT_DATE }} #${{ env.COMMIT_NUMBER_OF_DAY }}> " +
+                        "(<https://github.com/${{ github.repository }}/commit/${{ github.sha }}|" + "${{ github.sha }}".substring(0, 10) + ">) " +
+                        "of ${{ github.repository }}@" + "${{ github.ref }}".split('/').reverse()[0] + " by ${{ github.event.head_commit.committer.name }}",
+                "color": "danger"
+              }]
+            }
+        env:
+          SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}
+        if: failure() && github.event_name == 'push'

Dockerfile

@@ -42,8 +42,7 @@ FROM golang:1.16-alpine AS build-env
 
 WORKDIR /go/src/tailscale
 
-COPY go.mod .
-COPY go.sum .
+COPY go.mod go.sum ./
 RUN go mod download
 
 COPY . .

VERSION.txt

@@ -1 +1 @@
-1.7.0
+1.13.0

api.md

@@ -367,10 +367,11 @@ Etag: "e0b2816b418b3f266309d94426ac7668ab3c1fa87798785bf82f1085cc2f6d9c"
 
 #### `POST /api/v2/tailnet/:tailnet/acl` - set ACL for a tailnet
 
-Sets the ACL for the given tailnet. HuJSON and JSON are both accepted inputs. An `If-Match` header can be set to avoid missed updates.
+Sets the ACL for the given domain.
+HuJSON and JSON are both accepted inputs.
+An `If-Match` header can be set to avoid missed updates.
 
-Returns error for invalid ACLs.
-Returns error if using an `If-Match` header and the ETag does not match.
+Returns the updated ACL in JSON or HuJSON according to the `Accept` header on success. Otherwise, errors are returned for incorrectly defined ACLs, ACLs with failing tests on attempted updates, and mismatched `If-Match` header and ETag.
 
 ##### Parameters
 
@@ -380,7 +381,17 @@ Returns error if using an `If-Match` header and the ETag does not match.
 `Accept` - Sets the return type of the updated ACL. Response is parsed `JSON` if `application/json` is explicitly named, otherwise HuJSON will be returned.
 
 ###### POST Body
-ACL JSON or HuJSON (see https://tailscale.com/kb/1018/acls)
+
+The POST body should be a JSON or [HuJSON](https://github.com/tailscale/hujson#hujson---human-json) formatted JSON object.
+An ACL policy may contain the following top-level properties:
+
+* `Groups` - Static groups of users which can be used for ACL rules.
+* `Hosts` - Hostname aliases to use in place of IP addresses or subnets.
+* `ACLs` - Access control lists.
+* `TagOwners` - Defines who is allowed to use which tags.
+* `Tests` - Run on ACL updates to check correct functionality of defined ACLs.
+
+See https://tailscale.com/kb/1018/acls for more information on those properties.
 
 ##### Example
 ```
@@ -411,7 +422,7 @@ curl 'https://api.tailscale.com/api/v2/tailnet/example.com/acl' \
 }'
 ```
 
-Response
+Response:
 ```
 // Example/default ACLs for unrestricted connections.
 {
@@ -436,23 +447,40 @@ Response
 }
 ```
 
+Failed test error response:
+```
+{
+    "message": "test(s) failed",
+    "data": [
+        {
+            "user": "user1@example.com",
+            "errors": [
+                "address \"user2@example.com:400\": want: Accept, got: Drop"
+            ]
+        }
+    ]
+}
+```
+
 <a name=tailnet-acl-preview-post></a>
 
 #### `POST /api/v2/tailnet/:tailnet/acl/preview` - preview rule matches on an ACL for a resource
+
 Determines what rules match for a user on an ACL without saving the ACL to the server.
 
 ##### Parameters
 
 ###### Query Parameters
-`user` - A user's email. The provided ACL is queried with this user to determine which rules match.
+`type` - can be 'user' or 'ipport'
+`previewFor` - if type=user, a user's email. If type=ipport, a IP address + port like "10.0.0.1:80".
+The provided ACL is queried with this paramater to determine which rules match.
 
 ###### POST Body
 ACL JSON or HuJSON (see https://tailscale.com/kb/1018/acls)
 
 ##### Example
 ```
-POST /api/v2/tailnet/example.com/acl/preiew
-curl 'https://api.tailscale.com/api/v2/tailnet/example.com/acl?user=user1@example.com' \
+curl 'https://api.tailscale.com/api/v2/tailnet/example.com/acl/preview?previewFor=user1@example.com&type=user' \
   -u "tskey-yourapikey123:" \
   --data-binary '// Example/default ACLs for unrestricted connections.
 {
@@ -477,7 +505,7 @@ curl 'https://api.tailscale.com/api/v2/tailnet/example.com/acl?user=user1@exampl
 }'
 ```
 
-Response
+Response:
 ```
 {"matches":[{"users":["*"],"ports":["*:*"],"lineNumber":19}],"user":"user1@example.com"}
 ```

build_dist.sh

@@ -11,6 +11,36 @@
 
 set -eu
 
-eval $(./version/version.sh)
+IFS=".$IFS" read -r major minor patch <VERSION.txt
+git_hash=$(git rev-parse HEAD)
+if ! git diff-index --quiet HEAD; then
+	git_hash="${git_hash}-dirty"
+fi
+base_hash=$(git rev-list --max-count=1 HEAD -- VERSION.txt)
+change_count=$(git rev-list --count HEAD "^$base_hash")
+short_hash=$(echo "$git_hash" | cut -c1-9)
 
-exec go build -tags xversion -ldflags "-X tailscale.com/version.Long=${VERSION_LONG} -X tailscale.com/version.Short=${VERSION_SHORT} -X tailscale.com/version.GitCommit=${VERSION_GIT_HASH}" "$@"
+if expr "$minor" : "[0-9]*[13579]$" >/dev/null; then
+	patch="$change_count"
+	change_suffix=""
+elif [ "$change_count" != "0" ]; then
+	change_suffix="-$change_count"
+else
+	change_suffix=""
+fi
+
+long_suffix="$change_suffix-t$short_hash"
+SHORT="$major.$minor.$patch"
+LONG="${SHORT}$long_suffix"
+GIT_HASH="$git_hash"
+
+if [ "$1" = "shellvars" ]; then
+	cat <<EOF
+VERSION_SHORT="$SHORT"
+VERSION_LONG="$LONG"
+VERSION_GIT_HASH="$GIT_HASH"
+EOF
+	exit 0
+fi
+
+exec go build -ldflags "-X tailscale.com/version.Long=${LONG} -X tailscale.com/version.Short=${SHORT} -X tailscale.com/version.GitCommit=${GIT_HASH}" "$@"

build_docker.sh

@@ -25,7 +25,7 @@
 
 set -eu
 
-eval $(./version/version.sh)
+eval $(./build_dist.sh shellvars)
 
 docker build \
   --build-arg VERSION_LONG=$VERSION_LONG \

client/tailscale/apitype/apitype.go

@@ -0,0 +1,29 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package apitype contains types for the Tailscale local API.
+package apitype
+
+import "tailscale.com/tailcfg"
+
+// WhoIsResponse is the JSON type returned by tailscaled debug server's /whois?ip=$IP handler.
+type WhoIsResponse struct {
+	Node        *tailcfg.Node
+	UserProfile *tailcfg.UserProfile
+}
+
+// FileTarget is a node to which files can be sent, and the PeerAPI
+// URL base to do so via.
+type FileTarget struct {
+	Node *tailcfg.Node
+
+	// PeerAPI is the http://ip:port URL base of the node's peer API,
+	// without any path (not even a single slash).
+	PeerAPIURL string
+}
+
+type WaitingFile struct {
+	Name string
+	Size int64
+}

client/tailscale/tailscale.go

@@ -6,6 +6,7 @@
 package tailscale
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
@@ -18,6 +19,8 @@ import (
 	"strconv"
 	"strings"
 
+	"tailscale.com/client/tailscale/apitype"
+	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/paths"
 	"tailscale.com/safesocket"
@@ -64,9 +67,22 @@ func DoLocalRequest(req *http.Request) (*http.Response, error) {
 	return tsClient.Do(req)
 }
 
-// WhoIs returns the owner of the remoteAddr, which must be an IP or IP:port.
-func WhoIs(ctx context.Context, remoteAddr string) (*tailcfg.WhoIsResponse, error) {
-	req, err := http.NewRequestWithContext(ctx, "GET", "http://local-tailscaled.sock/localapi/v0/whois?addr="+url.QueryEscape(remoteAddr), nil)
+type errorJSON struct {
+	Error string
+}
+
+// bestError returns either err, or if body contains a valid JSON
+// object of type errorJSON, its non-empty error body.
+func bestError(err error, body []byte) error {
+	var j errorJSON
+	if err := json.Unmarshal(body, &j); err == nil && j.Error != "" {
+		return errors.New(j.Error)
+	}
+	return err
+}
+
+func send(ctx context.Context, method, path string, wantStatus int, body io.Reader) ([]byte, error) {
+	req, err := http.NewRequestWithContext(ctx, method, "http://local-tailscaled.sock"+path, body)
 	if err != nil {
 		return nil, err
 	}
@@ -75,60 +91,48 @@ func WhoIs(ctx context.Context, remoteAddr string) (*tailcfg.WhoIsResponse, erro
 		return nil, err
 	}
 	defer res.Body.Close()
-	slurp, _ := ioutil.ReadAll(res.Body)
-	if res.StatusCode != 200 {
-		return nil, fmt.Errorf("HTTP %s: %s", res.Status, slurp)
+	slurp, err := ioutil.ReadAll(res.Body)
+	if err != nil {
+		return nil, err
 	}
-	r := new(tailcfg.WhoIsResponse)
-	if err := json.Unmarshal(slurp, r); err != nil {
-		if max := 200; len(slurp) > max {
-			slurp = slurp[:max]
+	if res.StatusCode != wantStatus {
+		err := fmt.Errorf("HTTP %s: %s (expected %v)", res.Status, slurp, wantStatus)
+		return nil, bestError(err, slurp)
+	}
+	return slurp, nil
+}
+
+func get200(ctx context.Context, path string) ([]byte, error) {
+	return send(ctx, "GET", path, 200, nil)
+}
+
+// WhoIs returns the owner of the remoteAddr, which must be an IP or IP:port.
+func WhoIs(ctx context.Context, remoteAddr string) (*apitype.WhoIsResponse, error) {
+	body, err := get200(ctx, "/localapi/v0/whois?addr="+url.QueryEscape(remoteAddr))
+	if err != nil {
+		return nil, err
+	}
+	r := new(apitype.WhoIsResponse)
+	if err := json.Unmarshal(body, r); err != nil {
+		if max := 200; len(body) > max {
+			body = append(body[:max], "..."...)
 		}
-		return nil, fmt.Errorf("failed to parse JSON WhoIsResponse from %q", slurp)
+		return nil, fmt.Errorf("failed to parse JSON WhoIsResponse from %q", body)
 	}
 	return r, nil
 }
 
 // Goroutines returns a dump of the Tailscale daemon's current goroutines.
 func Goroutines(ctx context.Context) ([]byte, error) {
-	req, err := http.NewRequestWithContext(ctx, "GET", "http://local-tailscaled.sock/localapi/v0/goroutines", nil)
-	if err != nil {
-		return nil, err
-	}
-	res, err := DoLocalRequest(req)
-	if err != nil {
-		return nil, err
-	}
-	defer res.Body.Close()
-	body, err := ioutil.ReadAll(res.Body)
-	if err != nil {
-		return nil, err
-	}
-	if res.StatusCode != 200 {
-		return nil, fmt.Errorf("HTTP %s: %s", res.Status, body)
-	}
-	return body, nil
+	return get200(ctx, "/localapi/v0/goroutines")
 }
 
 // BugReport logs and returns a log marker that can be shared by the user with support.
 func BugReport(ctx context.Context, note string) (string, error) {
-	u := fmt.Sprintf("http://local-tailscaled.sock/localapi/v0/bugreport?note=%s", url.QueryEscape(note))
-	req, err := http.NewRequestWithContext(ctx, "POST", u, nil)
+	body, err := send(ctx, "POST", "/localapi/v0/bugreport?note="+url.QueryEscape(note), 200, nil)
 	if err != nil {
 		return "", err
 	}
-	res, err := DoLocalRequest(req)
-	if err != nil {
-		return "", err
-	}
-	defer res.Body.Close()
-	body, err := ioutil.ReadAll(res.Body)
-	if err != nil {
-		return "", err
-	}
-	if res.StatusCode != 200 {
-		return "", fmt.Errorf("HTTP %s: %s", res.Status, body)
-	}
 	return strings.TrimSpace(string(body)), nil
 }
 
@@ -143,67 +147,32 @@ func StatusWithoutPeers(ctx context.Context) (*ipnstate.Status, error) {
 }
 
 func status(ctx context.Context, queryString string) (*ipnstate.Status, error) {
-	req, err := http.NewRequestWithContext(ctx, "GET", "http://local-tailscaled.sock/localapi/v0/status"+queryString, nil)
+	body, err := get200(ctx, "/localapi/v0/status"+queryString)
 	if err != nil {
 		return nil, err
 	}
-	res, err := DoLocalRequest(req)
-	if err != nil {
-		return nil, err
-	}
-	defer res.Body.Close()
-	if res.StatusCode != 200 {
-		body, _ := ioutil.ReadAll(res.Body)
-		return nil, fmt.Errorf("HTTP %s: %s", res.Status, body)
-	}
 	st := new(ipnstate.Status)
-	if err := json.NewDecoder(res.Body).Decode(st); err != nil {
+	if err := json.Unmarshal(body, st); err != nil {
 		return nil, err
 	}
 	return st, nil
 }
 
-type WaitingFile struct {
-	Name string
-	Size int64
-}
-
-func WaitingFiles(ctx context.Context) ([]WaitingFile, error) {
-	req, err := http.NewRequestWithContext(ctx, "GET", "http://local-tailscaled.sock/localapi/v0/files/", nil)
+func WaitingFiles(ctx context.Context) ([]apitype.WaitingFile, error) {
+	body, err := get200(ctx, "/localapi/v0/files/")
 	if err != nil {
 		return nil, err
 	}
-	res, err := DoLocalRequest(req)
-	if err != nil {
-		return nil, err
-	}
-	defer res.Body.Close()
-	if res.StatusCode != 200 {
-		body, _ := ioutil.ReadAll(res.Body)
-		return nil, fmt.Errorf("HTTP %s: %s", res.Status, body)
-	}
-	var wfs []WaitingFile
-	if err := json.NewDecoder(res.Body).Decode(&wfs); err != nil {
+	var wfs []apitype.WaitingFile
+	if err := json.Unmarshal(body, &wfs); err != nil {
 		return nil, err
 	}
 	return wfs, nil
 }
 
 func DeleteWaitingFile(ctx context.Context, baseName string) error {
-	req, err := http.NewRequestWithContext(ctx, "DELETE", "http://local-tailscaled.sock/localapi/v0/files/"+url.PathEscape(baseName), nil)
-	if err != nil {
-		return err
-	}
-	res, err := DoLocalRequest(req)
-	if err != nil {
-		return err
-	}
-	defer res.Body.Close()
-	if res.StatusCode != http.StatusNoContent {
-		body, _ := ioutil.ReadAll(res.Body)
-		return fmt.Errorf("expected 204 No Content; got HTTP %s: %s", res.Status, body)
-	}
-	return nil
+	_, err := send(ctx, "DELETE", "/localapi/v0/files/"+url.PathEscape(baseName), http.StatusNoContent, nil)
+	return err
 }
 
 func GetWaitingFile(ctx context.Context, baseName string) (rc io.ReadCloser, size int64, err error) {
@@ -227,25 +196,27 @@ func GetWaitingFile(ctx context.Context, baseName string) (rc io.ReadCloser, siz
 	return res.Body, res.ContentLength, nil
 }
 
+func FileTargets(ctx context.Context) ([]apitype.FileTarget, error) {
+	body, err := get200(ctx, "/localapi/v0/file-targets")
+	if err != nil {
+		return nil, err
+	}
+	var fts []apitype.FileTarget
+	if err := json.Unmarshal(body, &fts); err != nil {
+		return nil, fmt.Errorf("invalid JSON: %w", err)
+	}
+	return fts, nil
+}
+
 func CheckIPForwarding(ctx context.Context) error {
-	req, err := http.NewRequestWithContext(ctx, "GET", "http://local-tailscaled.sock/localapi/v0/check-ip-forwarding", nil)
+	body, err := get200(ctx, "/localapi/v0/check-ip-forwarding")
 	if err != nil {
 		return err
 	}
-	res, err := DoLocalRequest(req)
-	if err != nil {
-		return err
-	}
-	defer res.Body.Close()
-	if res.StatusCode != 200 {
-		body, _ := ioutil.ReadAll(res.Body)
-		res.Body.Close()
-		return fmt.Errorf("HTTP %s: %s", res.Status, body)
-	}
 	var jres struct {
 		Warning string
 	}
-	if err := json.NewDecoder(res.Body).Decode(&jres); err != nil {
+	if err := json.Unmarshal(body, &jres); err != nil {
 		return fmt.Errorf("invalid JSON from check-ip-forwarding: %w", err)
 	}
 	if jres.Warning != "" {
@@ -253,3 +224,72 @@ func CheckIPForwarding(ctx context.Context) error {
 	}
 	return nil
 }
+
+func GetPrefs(ctx context.Context) (*ipn.Prefs, error) {
+	body, err := get200(ctx, "/localapi/v0/prefs")
+	if err != nil {
+		return nil, err
+	}
+	var p ipn.Prefs
+	if err := json.Unmarshal(body, &p); err != nil {
+		return nil, fmt.Errorf("invalid prefs JSON: %w", err)
+	}
+	return &p, nil
+}
+
+func EditPrefs(ctx context.Context, mp *ipn.MaskedPrefs) (*ipn.Prefs, error) {
+	mpj, err := json.Marshal(mp)
+	if err != nil {
+		return nil, err
+	}
+	body, err := send(ctx, "PATCH", "/localapi/v0/prefs", http.StatusOK, bytes.NewReader(mpj))
+	if err != nil {
+		return nil, err
+	}
+	var p ipn.Prefs
+	if err := json.Unmarshal(body, &p); err != nil {
+		return nil, fmt.Errorf("invalid prefs JSON: %w", err)
+	}
+	return &p, nil
+}
+
+func Logout(ctx context.Context) error {
+	_, err := send(ctx, "POST", "/localapi/v0/logout", http.StatusNoContent, nil)
+	return err
+}
+
+// SetDNS adds a DNS TXT record for the given domain name, containing
+// the provided TXT value. The intended use case is answering
+// LetsEncrypt/ACME dns-01 challenges.
+//
+// The control plane will only permit SetDNS requests with very
+// specific names and values. The name should be
+// "_acme-challenge." + your node's MagicDNS name. It's expected that
+// clients cache the certs from LetsEncrypt (or whichever CA is
+// providing them) and only request new ones as needed; the control plane
+// rate limits SetDNS requests.
+//
+// This is a low-level interface; it's expected that most Tailscale
+// users use a higher level interface to getting/using TLS
+// certificates.
+func SetDNS(ctx context.Context, name, value string) error {
+	v := url.Values{}
+	v.Set("name", name)
+	v.Set("value", value)
+	_, err := send(ctx, "POST", "/localapi/v0/set-dns?"+v.Encode(), 200, nil)
+	return err
+}
+
+// CurrentDERPMap returns the current DERPMap that is being used by the local tailscaled.
+// It is intended to be used with netcheck to see availability of DERPs.
+func CurrentDERPMap(ctx context.Context) (*tailcfg.DERPMap, error) {
+	var derpMap tailcfg.DERPMap
+	res, err := send(ctx, "GET", "/localapi/v0/derpmap", 200, nil)
+	if err != nil {
+		return nil, err
+	}
+	if err = json.Unmarshal(res, &derpMap); err != nil {
+		return nil, fmt.Errorf("invalid derp map json: %w", err)
+	}
+	return &derpMap, nil
+}

cmd/addlicense/main.go

@@ -0,0 +1,77 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Program addlicense adds a license header to a file.
+// It is intended for use with 'go generate',
+// so it has a slightly weird usage.
+package main
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"os/exec"
+)
+
+var (
+	year = flag.Int("year", 0, "copyright year")
+	file = flag.String("file", "", "file to modify")
+)
+
+func usage() {
+	fmt.Fprintf(os.Stderr, `
+usage: addlicense -year YEAR -file FILE <subcommand args...>
+`[1:])
+
+	flag.PrintDefaults()
+	fmt.Fprintf(os.Stderr, `
+addlicense adds a Tailscale license to the beginning of file,
+using year as the copyright year.
+
+It is intended for use with 'go generate', so it also runs a subcommand,
+which presumably creates the file.
+
+Sample usage:
+
+addlicense -year 2021 -file pull_strings.go stringer -type=pull
+`[1:])
+	os.Exit(2)
+}
+
+func main() {
+	flag.Usage = usage
+	flag.Parse()
+	if len(flag.Args()) == 0 {
+		flag.Usage()
+	}
+	cmd := exec.Command(flag.Arg(0), flag.Args()[1:]...)
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+	err := cmd.Run()
+	check(err)
+	b, err := os.ReadFile(*file)
+	check(err)
+	f, err := os.OpenFile(*file, os.O_TRUNC|os.O_WRONLY, 0644)
+	check(err)
+	_, err = fmt.Fprintf(f, license, *year)
+	check(err)
+	_, err = f.Write(b)
+	check(err)
+	err = f.Close()
+	check(err)
+}
+
+func check(err error) {
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+}
+
+var license = `
+// Copyright (c) %d Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+`[1:]

cmd/cloner/cloner.go

@@ -246,7 +246,9 @@ func gen(buf *bytes.Buffer, imports map[string]struct{}, name string, typ *types
 					writef("\t\tdst.%s[k] = append([]%s{}, src.%s[k]...)", fname, n, fname)
 					writef("\t}")
 				} else if containsPointers(ft.Elem()) {
-					writef("\t\t" + `panic("TODO map value pointers")`)
+					writef("\tfor k, v := range src.%s {", fname)
+					writef("\t\tdst.%s[k] = v.Clone()", fname)
+					writef("\t}")
 				} else {
 					writef("\tfor k, v := range src.%s {", fname)
 					writef("\t\tdst.%s[k] = v", fname)

cmd/derper/derper.go

@@ -12,8 +12,6 @@ import (
 	"errors"
 	"expvar"
 	"flag"
-	"fmt"
-	"html"
 	"io"
 	"io/ioutil"
 	"log"
@@ -35,7 +33,6 @@ import (
 	"tailscale.com/tsweb"
 	"tailscale.com/types/key"
 	"tailscale.com/types/wgkey"
-	"tailscale.com/version"
 )
 
 var (
@@ -49,6 +46,7 @@ var (
 	meshPSKFile   = flag.String("mesh-psk-file", defaultMeshPSKFile(), "if non-empty, path to file containing the mesh pre-shared key file. It should contain some hex string; whitespace is trimmed.")
 	meshWith      = flag.String("mesh-with", "", "optional comma-separated list of hostnames to mesh with; the server's own hostname can be in the list")
 	bootstrapDNS  = flag.String("bootstrap-dns-names", "", "optional comma-separated list of hostnames to make available at /bootstrap-dns")
+	verifyClients = flag.Bool("verify-clients", false, "verify clients to this DERP server through a local tailscaled instance.")
 )
 
 type config struct {
@@ -60,7 +58,12 @@ func loadConfig() config {
 		return config{PrivateKey: mustNewKey()}
 	}
 	if *configPath == "" {
-		log.Fatalf("derper: -c <config path> not specified")
+		if os.Getuid() == 0 {
+			*configPath = "/var/lib/derper/derper.key"
+		} else {
+			log.Fatalf("derper: -c <config path> not specified")
+		}
+		log.Printf("no config path specified; using %s", *configPath)
 	}
 	b, err := ioutil.ReadFile(*configPath)
 	switch {
@@ -125,6 +128,7 @@ func main() {
 	letsEncrypt := tsweb.IsProd443(*addr)
 
 	s := derp.NewServer(key.Private(cfg.PrivateKey), log.Printf)
+	s.SetVerifyClient(*verifyClients)
 
 	if *meshPSKFile != "" {
 		b, err := ioutil.ReadFile(*meshPSKFile)
@@ -143,8 +147,7 @@ func main() {
 	}
 	expvar.Publish("derp", s.ExpVar())
 
-	// Create our own mux so we don't expose /debug/ stuff to the world.
-	mux := tsweb.NewMux(debugHandler(s))
+	mux := http.NewServeMux()
 	mux.Handle("/derp", derphttp.Handler(s))
 	go refreshBootstrapDNSLoop()
 	mux.HandleFunc("/bootstrap-dns", handleBootstrapDNS)
@@ -164,6 +167,18 @@ func main() {
 			io.WriteString(w, "<p>Debug info at <a href='/debug/'>/debug/</a>.</p>\n")
 		}
 	}))
+	debug := tsweb.Debugger(mux)
+	debug.KV("TLS hostname", *hostname)
+	debug.KV("Mesh key", s.HasMeshKey())
+	debug.Handle("check", "Consistency check", http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		err := s.ConsistencyCheck()
+		if err != nil {
+			http.Error(w, err.Error(), 500)
+		} else {
+			io.WriteString(w, "derp.Server ConsistencyCheck okay")
+		}
+	}))
+	debug.Handle("traffic", "Traffic check", http.HandlerFunc(s.ServeDebugTraffic))
 
 	if *runSTUN {
 		go serveSTUN()
@@ -217,39 +232,6 @@ func main() {
 	}
 }
 
-func debugHandler(s *derp.Server) http.Handler {
-	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.RequestURI == "/debug/check" {
-			err := s.ConsistencyCheck()
-			if err != nil {
-				http.Error(w, err.Error(), 500)
-			} else {
-				io.WriteString(w, "derp.Server ConsistencyCheck okay")
-			}
-			return
-		}
-		f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
-		f(`<html><body>
-<h1>DERP debug</h1>
-<ul>
-`)
-		f("<li><b>Hostname:</b> %v</li>\n", html.EscapeString(*hostname))
-		f("<li><b>Uptime:</b> %v</li>\n", tsweb.Uptime())
-		f("<li><b>Mesh Key:</b> %v</li>\n", s.HasMeshKey())
-		f("<li><b>Version:</b> %v</li>\n", html.EscapeString(version.Long))
-
-		f(`<li><a href="/debug/vars">/debug/vars</a> (Go)</li>
-   <li><a href="/debug/varz">/debug/varz</a> (Prometheus)</li>
-   <li><a href="/debug/pprof/">/debug/pprof/</a></li>
-   <li><a href="/debug/pprof/goroutine?debug=1">/debug/pprof/goroutine</a> (collapsed)</li>
-   <li><a href="/debug/pprof/goroutine?debug=2">/debug/pprof/goroutine</a> (full)</li>
-   <li><a href="/debug/check">/debug/check</a> internal consistency check</li>
-<ul>
-</html>
-`)
-	})
-}
-
 func serveSTUN() {
 	pc, err := net.ListenPacket("udp", ":3478")
 	if err != nil {

cmd/derper/mesh.go

@@ -9,7 +9,9 @@ import (
 	"errors"
 	"fmt"
 	"log"
+	"net"
 	"strings"
+	"time"
 
 	"tailscale.com/derp"
 	"tailscale.com/derp/derphttp"
@@ -39,6 +41,34 @@ func startMeshWithHost(s *derp.Server, host string) error {
 		return err
 	}
 	c.MeshKey = s.MeshKey()
+
+	// For meshed peers within a region, connect via VPC addresses.
+	c.SetURLDialer(func(ctx context.Context, network, addr string) (net.Conn, error) {
+		host, port, err := net.SplitHostPort(addr)
+		if err != nil {
+			return nil, err
+		}
+		var d net.Dialer
+		var r net.Resolver
+		if port == "443" && strings.HasSuffix(host, ".tailscale.com") {
+			base := strings.TrimSuffix(host, ".tailscale.com")
+			subCtx, cancel := context.WithTimeout(ctx, 2*time.Second)
+			defer cancel()
+			vpcHost := base + "-vpc.tailscale.com"
+			ips, _ := r.LookupIP(subCtx, "ip", vpcHost)
+			if len(ips) > 0 {
+				vpcAddr := net.JoinHostPort(ips[0].String(), port)
+				c, err := d.DialContext(subCtx, network, vpcAddr)
+				if err == nil {
+					log.Printf("connected to %v (%v) instead of %v", vpcHost, ips[0], base)
+					return c, nil
+				}
+				log.Printf("failed to connect to %v (%v): %v; trying non-VPC route", vpcHost, ips[0], err)
+			}
+		}
+		return d.DialContext(ctx, network, addr)
+	})
+
 	add := func(k key.Public) { s.AddPacketForwarder(k, c) }
 	remove := func(k key.Public) { s.RemovePacketForwarder(k, c) }
 	go c.RunWatchConnectionLoop(context.Background(), s.PublicKey(), logf, add, remove)

cmd/derpprobe/derpprobe.go

@@ -0,0 +1,354 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// The derpprobe binary probes derpers.
+package main // import "tailscale.com/cmd/derper/derpprobe"
+
+import (
+	"bytes"
+	"context"
+	crand "crypto/rand"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"html"
+	"io"
+	"log"
+	"net/http"
+	"sort"
+	"sync"
+	"time"
+
+	"tailscale.com/derp"
+	"tailscale.com/derp/derphttp"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/key"
+)
+
+var (
+	derpMapURL = flag.String("derp-map", "https://login.tailscale.com/derpmap/default", "URL to DERP map (https:// or file://)")
+	listen     = flag.String("listen", ":8030", "HTTP listen address")
+)
+
+var (
+	mu            sync.Mutex
+	state         = map[nodePair]pairStatus{}
+	lastDERPMap   *tailcfg.DERPMap
+	lastDERPMapAt time.Time
+)
+
+func main() {
+	flag.Parse()
+	go probeLoop()
+	log.Fatal(http.ListenAndServe(*listen, http.HandlerFunc(serve)))
+}
+
+type overallStatus struct {
+	good, bad []string
+}
+
+func (st *overallStatus) addBadf(format string, a ...interface{}) {
+	st.bad = append(st.bad, fmt.Sprintf(format, a...))
+}
+
+func (st *overallStatus) addGoodf(format string, a ...interface{}) {
+	st.good = append(st.good, fmt.Sprintf(format, a...))
+}
+
+func getOverallStatus() (o overallStatus) {
+	mu.Lock()
+	defer mu.Unlock()
+	if lastDERPMap == nil {
+		o.addBadf("no DERP map")
+		return
+	}
+	now := time.Now()
+	if age := now.Sub(lastDERPMapAt); age > time.Minute {
+		o.addBadf("DERPMap hasn't been successfully refreshed in %v", age.Round(time.Second))
+	}
+	for _, reg := range sortedRegions(lastDERPMap) {
+		for _, from := range reg.Nodes {
+			for _, to := range reg.Nodes {
+				pair := nodePair{from.Name, to.Name}
+				st, ok := state[pair]
+				age := now.Sub(st.at).Round(time.Second)
+				switch {
+				case !ok:
+					o.addBadf("no state for %v", pair)
+				case st.err != nil:
+					o.addBadf("%v: %v", pair, st.err)
+				case age > 90*time.Second:
+					o.addBadf("%v: update is %v old", pair, age)
+				default:
+					o.addGoodf("%v: %v, %v ago", pair, st.latency.Round(time.Millisecond), age)
+				}
+			}
+		}
+	}
+	return
+}
+
+func serve(w http.ResponseWriter, r *http.Request) {
+	st := getOverallStatus()
+	summary := "All good"
+	if len(st.bad) > 0 {
+		w.WriteHeader(500)
+		summary = fmt.Sprintf("%d problems", len(st.bad))
+	}
+	io.WriteString(w, "<html><head><style>.bad { font-weight: bold; color: #700; }</style></head>\n")
+	fmt.Fprintf(w, "<body><h1>derp probe</h1>\n%s:<ul>", summary)
+	for _, s := range st.bad {
+		fmt.Fprintf(w, "<li class=bad>%s</li>\n", html.EscapeString(s))
+	}
+	for _, s := range st.good {
+		fmt.Fprintf(w, "<li>%s</li>\n", html.EscapeString(s))
+	}
+	io.WriteString(w, "</ul></body></html>\n")
+}
+
+func sortedRegions(dm *tailcfg.DERPMap) []*tailcfg.DERPRegion {
+	ret := make([]*tailcfg.DERPRegion, 0, len(dm.Regions))
+	for _, r := range dm.Regions {
+		ret = append(ret, r)
+	}
+	sort.Slice(ret, func(i, j int) bool { return ret[i].RegionID < ret[j].RegionID })
+	return ret
+}
+
+type nodePair struct {
+	from, to string // DERPNode.Name
+}
+
+func (p nodePair) String() string { return fmt.Sprintf("(%s→%s)", p.from, p.to) }
+
+type pairStatus struct {
+	err     error
+	latency time.Duration
+	at      time.Time
+}
+
+func setDERPMap(dm *tailcfg.DERPMap) {
+	mu.Lock()
+	defer mu.Unlock()
+	lastDERPMap = dm
+	lastDERPMapAt = time.Now()
+}
+
+func setState(p nodePair, latency time.Duration, err error) {
+	mu.Lock()
+	defer mu.Unlock()
+	st := pairStatus{
+		err:     err,
+		latency: latency,
+		at:      time.Now(),
+	}
+	state[p] = st
+	if err != nil {
+		log.Printf("%+v error: %v", p, err)
+	} else {
+		log.Printf("%+v: %v", p, latency.Round(time.Millisecond))
+	}
+}
+
+func probeLoop() {
+	ticker := time.NewTicker(15 * time.Second)
+	for {
+		err := probe()
+		if err != nil {
+			log.Printf("probe: %v", err)
+		}
+		<-ticker.C
+	}
+}
+
+func probe() error {
+	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	defer cancel()
+	dm, err := getDERPMap(ctx)
+	if err != nil {
+		return err
+	}
+
+	var wg sync.WaitGroup
+	wg.Add(len(dm.Regions))
+	for _, reg := range dm.Regions {
+		reg := reg
+		go func() {
+			defer wg.Done()
+			for _, from := range reg.Nodes {
+				for _, to := range reg.Nodes {
+					latency, err := probeNodePair(ctx, dm, from, to)
+					setState(nodePair{from.Name, to.Name}, latency, err)
+				}
+			}
+		}()
+	}
+
+	wg.Wait()
+	return ctx.Err()
+}
+
+func probeNodePair(ctx context.Context, dm *tailcfg.DERPMap, from, to *tailcfg.DERPNode) (latency time.Duration, err error) {
+	// The passed in context is a minute for the whole region. The
+	// idea is that each node pair in the region will be done
+	// serially and regularly in the future, reusing connections
+	// (at least in the happy path). For now they don't reuse
+	// connections and probe at most once every 15 seconds. We
+	// bound the duration of a single node pair within a region
+	// so one bad one can't starve others.
+	ctx, cancel := context.WithTimeout(ctx, 10*time.Second)
+	defer cancel()
+
+	fromc, err := newConn(ctx, dm, from)
+	if err != nil {
+		return 0, err
+	}
+	defer fromc.Close()
+	toc, err := newConn(ctx, dm, to)
+	if err != nil {
+		return 0, err
+	}
+	defer toc.Close()
+
+	// Wait a bit for from's node to hear about to existing on the
+	// other node in the region, in the case where the two nodes
+	// are different.
+	if from.Name != to.Name {
+		time.Sleep(100 * time.Millisecond) // pretty arbitrary
+	}
+
+	// Make a random packet
+	pkt := make([]byte, 8)
+	crand.Read(pkt)
+
+	t0 := time.Now()
+
+	// Send the random packet.
+	sendc := make(chan error, 1)
+	go func() {
+		sendc <- fromc.Send(toc.SelfPublicKey(), pkt)
+	}()
+	select {
+	case <-ctx.Done():
+		return 0, fmt.Errorf("timeout sending via %q: %w", from.Name, ctx.Err())
+	case err := <-sendc:
+		if err != nil {
+			return 0, fmt.Errorf("error sending via %q: %w", from.Name, err)
+		}
+	}
+
+	// Receive the random packet.
+	recvc := make(chan interface{}, 1) // either derp.ReceivedPacket or error
+	go func() {
+		for {
+			m, err := toc.Recv()
+			if err != nil {
+				recvc <- err
+				return
+			}
+			switch v := m.(type) {
+			case derp.ReceivedPacket:
+				recvc <- v
+			default:
+				log.Printf("%v: ignoring Recv frame type %T", to.Name, v)
+				// Loop.
+			}
+		}
+	}()
+	select {
+	case <-ctx.Done():
+		return 0, fmt.Errorf("timeout receiving from %q: %w", to.Name, ctx.Err())
+	case v := <-recvc:
+		if err, ok := v.(error); ok {
+			return 0, fmt.Errorf("error receiving from %q: %w", to.Name, err)
+		}
+		p := v.(derp.ReceivedPacket)
+		if p.Source != fromc.SelfPublicKey() {
+			return 0, fmt.Errorf("got data packet from unexpected source, %v", p.Source)
+		}
+		if !bytes.Equal(p.Data, pkt) {
+			return 0, fmt.Errorf("unexpected data packet %q", p.Data)
+		}
+	}
+	return time.Since(t0), nil
+}
+
+func newConn(ctx context.Context, dm *tailcfg.DERPMap, n *tailcfg.DERPNode) (*derphttp.Client, error) {
+	priv := key.NewPrivate()
+	dc := derphttp.NewRegionClient(priv, log.Printf, func() *tailcfg.DERPRegion {
+		rid := n.RegionID
+		return &tailcfg.DERPRegion{
+			RegionID:   rid,
+			RegionCode: fmt.Sprintf("%s-%s", dm.Regions[rid].RegionCode, n.Name),
+			RegionName: dm.Regions[rid].RegionName,
+			Nodes:      []*tailcfg.DERPNode{n},
+		}
+	})
+	dc.IsProber = true
+	err := dc.Connect(ctx)
+	if err != nil {
+		return nil, err
+	}
+	errc := make(chan error, 1)
+	go func() {
+		m, err := dc.Recv()
+		if err != nil {
+			errc <- err
+			return
+		}
+		switch m.(type) {
+		case derp.ServerInfoMessage:
+			errc <- nil
+		default:
+			errc <- fmt.Errorf("unexpected first message type %T", errc)
+		}
+	}()
+	select {
+	case err := <-errc:
+		if err != nil {
+			go dc.Close()
+			return nil, err
+		}
+	case <-ctx.Done():
+		go dc.Close()
+		return nil, fmt.Errorf("timeout waiting for ServerInfoMessage: %w", ctx.Err())
+	}
+	return dc, nil
+}
+
+var httpOrFileClient = &http.Client{Transport: httpOrFileTransport()}
+
+func httpOrFileTransport() http.RoundTripper {
+	tr := http.DefaultTransport.(*http.Transport).Clone()
+	tr.RegisterProtocol("file", http.NewFileTransport(http.Dir("/")))
+	return tr
+}
+
+func getDERPMap(ctx context.Context) (*tailcfg.DERPMap, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", *derpMapURL, nil)
+	if err != nil {
+		return nil, err
+	}
+	res, err := httpOrFileClient.Do(req)
+	if err != nil {
+		mu.Lock()
+		defer mu.Unlock()
+		if lastDERPMap != nil && time.Since(lastDERPMapAt) < 10*time.Minute {
+			// Assume that control is restarting and use
+			// the same one for a bit.
+			return lastDERPMap, nil
+		}
+		return nil, err
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		return nil, fmt.Errorf("fetching %s: %s", *derpMapURL, res.Status)
+	}
+	dm := new(tailcfg.DERPMap)
+	if err := json.NewDecoder(res.Body).Decode(dm); err != nil {
+		return nil, fmt.Errorf("decoding %s JSON: %v", *derpMapURL, err)
+	}
+	setDERPMap(dm)
+	return dm, nil
+}

cmd/hello/hello.go

@@ -18,7 +18,7 @@ import (
 	"strings"
 
 	"tailscale.com/client/tailscale"
-	"tailscale.com/tailcfg"
+	"tailscale.com/client/tailscale/apitype"
 )
 
 var (
@@ -107,18 +107,18 @@ type tmplData struct {
 	IP            string // "100.2.3.4"
 }
 
-func tailscaleIP(who *tailcfg.WhoIsResponse) string {
+func tailscaleIP(who *apitype.WhoIsResponse) string {
 	if who == nil {
 		return ""
 	}
 	for _, nodeIP := range who.Node.Addresses {
-		if nodeIP.IP.Is4() && nodeIP.IsSingleIP() {
-			return nodeIP.IP.String()
+		if nodeIP.IP().Is4() && nodeIP.IsSingleIP() {
+			return nodeIP.IP().String()
 		}
 	}
 	for _, nodeIP := range who.Node.Addresses {
 		if nodeIP.IsSingleIP() {
-			return nodeIP.IP.String()
+			return nodeIP.IP().String()
 		}
 	}
 	return ""

cmd/microproxy/microproxy.go

@@ -55,9 +55,13 @@ func main() {
 		log.Fatalf("Couldn't parse URL %q: %v", *goVarsURL, err)
 	}
 
-	mux := tsweb.NewMux(http.HandlerFunc(debugHandler))
+	mux := http.NewServeMux()
+	tsweb.Debugger(mux) // registers /debug/*
 	mux.Handle("/metrics", tsweb.Protected(proxy))
-	mux.Handle("/varz", tsweb.Protected(tsweb.StdHandler(&goVarsHandler{*goVarsURL}, log.Printf)))
+	mux.Handle("/varz", tsweb.Protected(tsweb.StdHandler(&goVarsHandler{*goVarsURL}, tsweb.HandlerOptions{
+		Quiet200s: true,
+		Logf:      log.Printf,
+	})))
 
 	ch := &certHolder{
 		hostname: *hostname,
@@ -167,23 +171,3 @@ func (c *certHolder) loadLocked() error {
 	c.loaded = time.Now()
 	return nil
 }
-
-// debugHandler serves a page with links to tsweb-managed debug URLs
-// at /debug/.
-func debugHandler(w http.ResponseWriter, r *http.Request) {
-	f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
-	f(`<html><body>
-<h1>microproxy debug</h1>
-<ul>
-`)
-	f("<li><b>Hostname:</b> %v</li>\n", *hostname)
-	f("<li><b>Uptime:</b> %v</li>\n", tsweb.Uptime())
-	f(`<li><a href="/debug/vars">/debug/vars</a> (Go)</li>
-   <li><a href="/debug/varz">/debug/varz</a> (Prometheus)</li>
-   <li><a href="/debug/pprof/">/debug/pprof/</a></li>
-   <li><a href="/debug/pprof/goroutine?debug=1">/debug/pprof/goroutine</a> (collapsed)</li>
-   <li><a href="/debug/pprof/goroutine?debug=2">/debug/pprof/goroutine</a> (full)</li>
-<ul>
-</html>
-`)
-}

cmd/mkpkg/main.go

@@ -21,6 +21,9 @@ import (
 // into a map of filePathOnDisk -> filePathInPackage.
 func parseFiles(s string) (map[string]string, error) {
 	ret := map[string]string{}
+	if len(s) == 0 {
+		return ret, nil
+	}
 	for _, f := range strings.Split(s, ",") {
 		fs := strings.Split(f, ":")
 		if len(fs) != 2 {

cmd/speedtest/speedtest.go

@@ -0,0 +1,121 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Program speedtest provides the speedtest command. The reason to keep it separate from
+// the normal tailscale cli is because it is not yet ready to go in the tailscale binary.
+// It will be included in the tailscale cli after it has been added to tailscaled.
+
+// Example usage for client command: go run cmd/speedtest -host 127.0.0.1:20333 -t 5s
+// This will connect to the server on 127.0.0.1:20333 and start a 5 second download speedtest.
+// Example usage for server command: go run cmd/speedtest -s -host :20333
+// This will start a speedtest server on port 20333.
+package main
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"net"
+	"os"
+	"strconv"
+	"text/tabwriter"
+	"time"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"tailscale.com/net/speedtest"
+)
+
+// Runs the speedtest command as a commandline program
+func main() {
+	args := os.Args[1:]
+	if err := speedtestCmd.Parse(args); err != nil {
+		fmt.Fprintln(os.Stderr, err.Error())
+		os.Exit(1)
+	}
+
+	err := speedtestCmd.Run(context.Background())
+	if errors.Is(err, flag.ErrHelp) {
+		fmt.Fprintln(os.Stderr, speedtestCmd.ShortUsage)
+		os.Exit(2)
+	}
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err.Error())
+		os.Exit(1)
+	}
+}
+
+// speedtestCmd is the root command. It runs either the server or client depending on the
+// flags passed to it.
+var speedtestCmd = &ffcli.Command{
+	Name:       "speedtest",
+	ShortUsage: "speedtest [-host <host:port>] [-s] [-r] [-t <test duration>]",
+	ShortHelp:  "Run a speed test",
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("speedtest", flag.ExitOnError)
+		fs.StringVar(&speedtestArgs.host, "host", ":20333", "host:port pair to connect to or listen on")
+		fs.DurationVar(&speedtestArgs.testDuration, "t", speedtest.DefaultDuration, "duration of the speed test")
+		fs.BoolVar(&speedtestArgs.runServer, "s", false, "run a speedtest server")
+		fs.BoolVar(&speedtestArgs.reverse, "r", false, "run in reverse mode (server sends, client receives)")
+		return fs
+	})(),
+	Exec: runSpeedtest,
+}
+
+var speedtestArgs struct {
+	host         string
+	testDuration time.Duration
+	runServer    bool
+	reverse      bool
+}
+
+func runSpeedtest(ctx context.Context, args []string) error {
+
+	if _, _, err := net.SplitHostPort(speedtestArgs.host); err != nil {
+		var addrErr *net.AddrError
+		if errors.As(err, &addrErr) && addrErr.Err == "missing port in address" {
+			// if no port is provided, append the default port
+			speedtestArgs.host = net.JoinHostPort(speedtestArgs.host, strconv.Itoa(speedtest.DefaultPort))
+		}
+	}
+
+	if speedtestArgs.runServer {
+		listener, err := net.Listen("tcp", speedtestArgs.host)
+		if err != nil {
+			return err
+		}
+
+		fmt.Printf("listening on %v\n", listener.Addr())
+
+		return speedtest.Serve(listener)
+	}
+
+	// Ensure the duration is within the allowed range
+	if speedtestArgs.testDuration < speedtest.MinDuration || speedtestArgs.testDuration > speedtest.MaxDuration {
+		return fmt.Errorf("test duration must be within %v and %v", speedtest.MinDuration, speedtest.MaxDuration)
+	}
+
+	dir := speedtest.Download
+	if speedtestArgs.reverse {
+		dir = speedtest.Upload
+	}
+
+	fmt.Printf("Starting a %s test with %s\n", dir, speedtestArgs.host)
+	results, err := speedtest.RunClient(dir, speedtestArgs.testDuration, speedtestArgs.host)
+	if err != nil {
+		return err
+	}
+
+	w := tabwriter.NewWriter(os.Stdout, 12, 0, 0, ' ', tabwriter.TabIndent)
+	fmt.Println("Results:")
+	fmt.Fprintln(w, "Interval\t\tTransfer\t\tBandwidth\t\t")
+	for _, r := range results {
+		if r.Total {
+			fmt.Fprintln(w, "-------------------------------------------------------------------------")
+		}
+		fmt.Fprintf(w, "%.2f-%.2f\tsec\t%.4f\tMBits\t%.4f\tMbits/sec\t\n", r.IntervalStart.Seconds(), r.IntervalEnd.Seconds(), r.MegaBits(), r.MBitsPerSecond())
+	}
+	w.Flush()
+	return nil
+}

cmd/tailscale/cli/admin.go

@@ -0,0 +1,155 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"strings"
+
+	"github.com/dsnet/golib/jsonfmt"
+	"github.com/peterbourgon/ff/v2/ffcli"
+)
+
+const tailscaleAPIURL = "https://api.tailscale.com/api"
+
+var adminCmd = &ffcli.Command{
+	Name:       "admin",
+	ShortUsage: "admin <subcommand> [command flags]",
+	ShortHelp:  "Administrate a tailnet",
+	LongHelp: strings.TrimSpace(`
+The "tailscale admin" command administrates a tailnet through the CLI.
+It is a wrapper over the RESTful API served at ` + tailscaleAPIURL + `.
+See https://github.com/tailscale/tailscale/blob/main/api.md for more information
+about the API itself.
+
+In order for the "admin" command to call the API, it needs an API key,
+which is specified by setting the TAILSCALE_API_KEY environment variable.
+Also, to easy usage, the tailnet to administrate can be specified through the
+TAILSCALE_NET_NAME environment variable, or specified with the -tailnet flag.
+
+Visit https://login.tailscale.com/admin/settings/authkeys in order to obtain
+an API key.
+`),
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("status", flag.ExitOnError)
+		// TODO(dsnet): Can we determine the default tailnet from what this
+		// device is currently part of? Alternatively, when add specific logic
+		// to handle auth keys, we can always associate a given key with a
+		// specific tailnet.
+		fs.StringVar(&adminArgs.tailnet, "tailnet", os.Getenv("TAILSCALE_NET_NAME"), "which tailnet to administrate")
+		return fs
+	})(),
+	// TODO(dsnet): Handle users, groups, dns.
+	Subcommands: []*ffcli.Command{{
+		Name:       "acl",
+		ShortUsage: "acl <subcommand> [command flags]",
+		ShortHelp:  "Manage the ACL for a tailnet",
+		// TODO(dsnet): Handle preview.
+		Subcommands: []*ffcli.Command{{
+			Name:       "get",
+			ShortUsage: "get",
+			ShortHelp:  "Downloads the HuJSON ACL file to stdout",
+			Exec:       checkAdminKey(runAdminACLGet),
+		}, {
+			Name:       "set",
+			ShortUsage: "set",
+			ShortHelp:  "Uploads the HuJSON ACL file from stdin",
+			Exec:       checkAdminKey(runAdminACLSet),
+		}},
+		Exec: runHelp,
+	}, {
+		Name:       "devices",
+		ShortUsage: "devices <subcommand> [command flags]",
+		ShortHelp:  "Manage devices in a tailnet",
+		Subcommands: []*ffcli.Command{{
+			Name:       "list",
+			ShortUsage: "list",
+			ShortHelp:  "List all devices in a tailnet",
+			Exec:       checkAdminKey(runAdminDevicesList),
+		}, {
+			Name:       "get",
+			ShortUsage: "get <id>",
+			ShortHelp:  "Get information about a specific device",
+			Exec:       checkAdminKey(runAdminDevicesGet),
+		}},
+		Exec: runHelp,
+	}},
+	Exec: runHelp,
+}
+
+var adminArgs struct {
+	tailnet string // which tailnet to operate upon
+}
+
+func checkAdminKey(f func(context.Context, string, []string) error) func(context.Context, []string) error {
+	return func(ctx context.Context, args []string) error {
+		// TODO(dsnet): We should have a subcommand or flag to manage keys.
+		// Use of an environment variable is a temporary hack.
+		key := os.Getenv("TAILSCALE_API_KEY")
+		if !strings.HasPrefix(key, "tskey-") {
+			return errors.New("no API key specified")
+		}
+		return f(ctx, key, args)
+	}
+}
+
+func runAdminACLGet(ctx context.Context, key string, args []string) error {
+	if len(args) > 0 {
+		return flag.ErrHelp
+	}
+	return adminCallAPI(ctx, key, http.MethodGet, "/v2/tailnet/"+adminArgs.tailnet+"/acl", nil, os.Stdout)
+}
+
+func runAdminACLSet(ctx context.Context, key string, args []string) error {
+	if len(args) > 0 {
+		return flag.ErrHelp
+	}
+	return adminCallAPI(ctx, key, http.MethodPost, "/v2/tailnet/"+adminArgs.tailnet+"/acl", os.Stdin, os.Stdout)
+}
+
+func runAdminDevicesList(ctx context.Context, key string, args []string) error {
+	if len(args) > 0 {
+		return flag.ErrHelp
+	}
+	return adminCallAPI(ctx, key, http.MethodGet, "/v2/tailnet/"+adminArgs.tailnet+"/devices", nil, os.Stdout)
+}
+
+func runAdminDevicesGet(ctx context.Context, key string, args []string) error {
+	if len(args) != 1 {
+		return flag.ErrHelp
+	}
+	return adminCallAPI(ctx, key, http.MethodGet, "/v2/device/"+args[0], nil, os.Stdout)
+}
+
+func adminCallAPI(ctx context.Context, key, method, path string, in io.Reader, out io.Writer) error {
+	req, err := http.NewRequestWithContext(ctx, method, tailscaleAPIURL+path, in)
+	req.SetBasicAuth(key, "")
+	if err != nil {
+		return fmt.Errorf("failed to create request: %w", err)
+	}
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("failed to send HTTP request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	b, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return fmt.Errorf("failed to receive HTTP response: %w", err)
+	}
+	b, err = jsonfmt.Format(b)
+	if err != nil {
+		return fmt.Errorf("failed to format JSON response: %w", err)
+	}
+	_, err = out.Write(b)
+	return err
+
+}

cmd/tailscale/cli/auth-redirect.html

@@ -0,0 +1,57 @@
+<html>
+<head>
+	<title>Redirecting...</title>
+	<style>
+		html,
+		body {
+			height: 100%;
+		}
+
+		html {
+			background-color: rgb(249, 247, 246);
+			font-family: ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol", "Noto Color Emoji";
+			line-height: 1.5;
+			-webkit-text-size-adjust: 100%;
+			-webkit-font-smoothing: antialiased;
+			-moz-osx-font-smoothing: grayscale;
+		}
+
+		body {
+			display: flex;
+			flex-direction: column;
+			align-items: center;
+			justify-content: center;
+		}
+
+		.spinner {
+			margin-bottom: 2rem;
+			border: 4px rgba(112, 110, 109, 0.5) solid;
+			border-left-color: transparent;
+			border-radius: 9999px;
+			width: 4rem;
+			height: 4rem;
+			-webkit-animation: spin 700ms linear infinite;
+      animation: spin 800ms linear infinite;
+		}
+
+		.label {
+			color: rgb(112, 110, 109);
+			padding-left: 0.4rem;
+		}
+
+		@-webkit-keyframes spin {
+			to {
+				transform: rotate(360deg);
+			}
+		}
+
+		@keyframes spin {
+			to {
+				transform: rotate(360deg);
+			}
+		}
+	</style>
+</head> <body>
+	<div class="spinner"></div>
+	<div class="label">Redirecting...</div>
+</body>

cmd/tailscale/cli/cli.go

@@ -8,13 +8,16 @@ package cli
 
 import (
 	"context"
+	"errors"
 	"flag"
 	"fmt"
+	"io"
 	"log"
 	"net"
 	"os"
 	"os/signal"
 	"runtime"
+	"strconv"
 	"strings"
 	"syscall"
 	"text/tabwriter"
@@ -24,24 +27,59 @@ import (
 	"tailscale.com/ipn"
 	"tailscale.com/paths"
 	"tailscale.com/safesocket"
+	"tailscale.com/syncs"
 )
 
 // ActLikeCLI reports whether a GUI application should act like the
 // CLI based on os.Args, GOOS, the context the process is running in
 // (pty, parent PID), etc.
 func ActLikeCLI() bool {
-	if len(os.Args) < 2 {
+	// This function is only used on macOS.
+	if runtime.GOOS != "darwin" {
 		return false
 	}
-	switch os.Args[1] {
-	case "up", "down", "status", "netcheck", "ping", "version",
-		"debug",
-		"-V", "--version", "-h", "--help":
+
+	// Escape hatch to let people force running the macOS
+	// GUI Tailscale binary as the CLI.
+	if v, _ := strconv.ParseBool(os.Getenv("TAILSCALE_BE_CLI")); v {
 		return true
 	}
+
+	// If our parent is launchd, we're definitely not
+	// being run as a CLI.
+	if os.Getppid() == 1 {
+		return false
+	}
+
+	// Xcode adds the -NSDocumentRevisionsDebugMode flag on execution.
+	// If present, we are almost certainly being run as a GUI.
+	for _, arg := range os.Args {
+		if arg == "-NSDocumentRevisionsDebugMode" {
+			return false
+		}
+	}
+
+	// Looking at the environment of the GUI Tailscale app (ps eww
+	// $PID), empirically none of these environment variables are
+	// present. But all or some of these should be present with
+	// Terminal.all and bash or zsh.
+	for _, e := range []string{
+		"SHLVL",
+		"TERM",
+		"TERM_PROGRAM",
+		"PS1",
+	} {
+		if os.Getenv(e) != "" {
+			return true
+		}
+	}
 	return false
 }
 
+func runHelp(context.Context, []string) error {
+	return flag.ErrHelp
+}
+
 // Run runs the CLI. The args do not include the binary name.
 func Run(args []string) error {
 	if len(args) == 1 && (args[0] == "-V" || args[0] == "--version") {
@@ -64,17 +102,19 @@ change in the future.
 		Subcommands: []*ffcli.Command{
 			upCmd,
 			downCmd,
+			logoutCmd,
+			adminCmd,
 			netcheckCmd,
 			ipCmd,
 			statusCmd,
 			pingCmd,
 			versionCmd,
 			webCmd,
-			pushCmd,
+			fileCmd,
 			bugReportCmd,
 		},
 		FlagSet:   rootfs,
-		Exec:      func(context.Context, []string) error { return flag.ErrHelp },
+		Exec:      runHelp,
 		UsageFunc: usageFunc,
 	}
 	for _, c := range rootCmd.Subcommands {
@@ -108,6 +148,8 @@ var rootArgs struct {
 	socket string
 }
 
+var gotSignal syncs.AtomicBool
+
 func connect(ctx context.Context) (net.Conn, *ipn.BackendClient, context.Context, context.CancelFunc) {
 	c, err := safesocket.Connect(rootArgs.socket, 41112)
 	if err != nil {
@@ -125,7 +167,14 @@ func connect(ctx context.Context) (net.Conn, *ipn.BackendClient, context.Context
 	go func() {
 		interrupt := make(chan os.Signal, 1)
 		signal.Notify(interrupt, syscall.SIGINT, syscall.SIGTERM)
-		<-interrupt
+		select {
+		case <-interrupt:
+		case <-ctx.Done():
+			// Context canceled elsewhere.
+			signal.Reset(syscall.SIGINT, syscall.SIGTERM)
+			return
+		}
+		gotSignal.Set(true)
 		c.Close()
 		cancel()
 	}()
@@ -135,19 +184,22 @@ func connect(ctx context.Context) (net.Conn, *ipn.BackendClient, context.Context
 }
 
 // pump receives backend messages on conn and pushes them into bc.
-func pump(ctx context.Context, bc *ipn.BackendClient, conn net.Conn) {
+func pump(ctx context.Context, bc *ipn.BackendClient, conn net.Conn) error {
 	defer conn.Close()
 	for ctx.Err() == nil {
 		msg, err := ipn.ReadMsg(conn)
 		if err != nil {
 			if ctx.Err() != nil {
-				return
+				return ctx.Err()
 			}
-			log.Printf("ReadMsg: %v\n", err)
-			break
+			if errors.Is(err, io.EOF) || errors.Is(err, net.ErrClosed) {
+				return fmt.Errorf("%w (tailscaled stopped running?)", err)
+			}
+			return err
 		}
 		bc.GotNotifyMsg(msg)
 	}
+	return ctx.Err()
 }
 
 func strSliceContains(ss []string, s string) bool {

cmd/tailscale/cli/cli_test.go

@@ -0,0 +1,798 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"bytes"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/google/go-cmp/cmp"
+	"inet.af/netaddr"
+	"tailscale.com/ipn"
+	"tailscale.com/ipn/ipnstate"
+	"tailscale.com/types/persist"
+	"tailscale.com/types/preftype"
+)
+
+// geese is a collection of gooses. It need not be complete.
+// But it should include anything handled specially (e.g. linux, windows)
+// and at least one thing that's not (darwin, freebsd).
+var geese = []string{"linux", "darwin", "windows", "freebsd"}
+
+// Test that checkForAccidentalSettingReverts's updateMaskedPrefsFromUpFlag can handle
+// all flags. This will panic if a new flag creeps in that's unhandled.
+//
+// Also, issue 1880: advertise-exit-node was being ignored. Verify that all flags cause an edit.
+func TestUpdateMaskedPrefsFromUpFlag(t *testing.T) {
+	for _, goos := range geese {
+		var upArgs upArgsT
+		fs := newUpFlagSet(goos, &upArgs)
+		fs.VisitAll(func(f *flag.Flag) {
+			mp := new(ipn.MaskedPrefs)
+			updateMaskedPrefsFromUpFlag(mp, f.Name)
+			got := mp.Pretty()
+			wantEmpty := preflessFlag(f.Name)
+			isEmpty := got == "MaskedPrefs{}"
+			if isEmpty != wantEmpty {
+				t.Errorf("flag %q created MaskedPrefs %s; want empty=%v", f.Name, got, wantEmpty)
+			}
+		})
+	}
+}
+
+func TestCheckForAccidentalSettingReverts(t *testing.T) {
+	tests := []struct {
+		name     string
+		flags    []string // argv to be parsed by FlagSet
+		curPrefs *ipn.Prefs
+
+		curExitNodeIP netaddr.IP
+		curUser       string // os.Getenv("USER") on the client side
+		goos          string // empty means "linux"
+
+		want string
+	}{
+		{
+			name:  "bare_up_means_up",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL:  ipn.DefaultControlURL,
+				WantRunning: false,
+				Hostname:    "foo",
+			},
+			want: "",
+		},
+		{
+			name:  "losing_hostname",
+			flags: []string{"--accept-dns"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      false,
+				Hostname:         "foo",
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+			},
+			want: accidentalUpPrefix + " --accept-dns --hostname=foo",
+		},
+		{
+			name:  "hostname_changing_explicitly",
+			flags: []string{"--hostname=bar"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+				Hostname:         "foo",
+			},
+			want: "",
+		},
+		{
+			name:  "hostname_changing_empty_explicitly",
+			flags: []string{"--hostname="},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+				Hostname:         "foo",
+			},
+			want: "",
+		},
+		{
+			// Issue 1725: "tailscale up --authkey=..." (or other non-empty flags) works from
+			// a fresh server's initial prefs.
+			name:     "up_with_default_prefs",
+			flags:    []string{"--authkey=foosdlkfjskdljf"},
+			curPrefs: ipn.NewPrefs(),
+			want:     "",
+		},
+		{
+			name:  "implicit_operator_change",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				OperatorUser:     "alice",
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			curUser: "eve",
+			want:    accidentalUpPrefix + " --hostname=foo --operator=alice",
+		},
+		{
+			name:  "implicit_operator_matches_shell_user",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				OperatorUser:     "alice",
+			},
+			curUser: "alice",
+			want:    "",
+		},
+		{
+			name:  "error_advertised_routes_exit_node_removed",
+			flags: []string{"--advertise-routes=10.0.42.0/24"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.42.0/24"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-routes=10.0.42.0/24 --advertise-exit-node",
+		},
+		{
+			name:  "advertised_routes_exit_node_removed_explicit",
+			flags: []string{"--advertise-routes=10.0.42.0/24", "--advertise-exit-node=false"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.42.0/24"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+			},
+			want: "",
+		},
+		{
+			name:  "advertised_routes_includes_the_0_routes", // but no --advertise-exit-node
+			flags: []string{"--advertise-routes=11.1.43.0/24,0.0.0.0/0,::/0"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.42.0/24"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+			},
+			want: "",
+		},
+		{
+			name:  "advertise_exit_node", // Issue 1859
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			want: "",
+		},
+		{
+			name:  "advertise_exit_node_over_existing_routes",
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
+		},
+		{
+			name:  "advertise_exit_node_over_existing_routes_and_exit_node",
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
+		},
+		{
+			name:  "exit_node_clearing", // Issue 1777
+			flags: []string{"--exit-node="},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				ExitNodeID: "fooID",
+			},
+			want: "",
+		},
+		{
+			name:  "remove_all_implicit",
+			flags: []string{"--force-reauth"},
+			curPrefs: &ipn.Prefs{
+				WantRunning:      true,
+				ControlURL:       ipn.DefaultControlURL,
+				RouteAll:         true,
+				AllowSingleHosts: false,
+				ExitNodeIP:       netaddr.MustParseIP("100.64.5.6"),
+				CorpDNS:          false,
+				ShieldsUp:        true,
+				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
+				Hostname:         "myhostname",
+				ForceDaemon:      true,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.0.0/16"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+				NetfilterMode: preftype.NetfilterNoDivert,
+				OperatorUser:  "alice",
+			},
+			curUser: "eve",
+			want:    accidentalUpPrefix + " --force-reauth --accept-dns=false --accept-routes --advertise-exit-node --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --host-routes=false --hostname=myhostname --netfilter-mode=nodivert --operator=alice --shields-up",
+		},
+		{
+			name:  "remove_all_implicit_except_hostname",
+			flags: []string{"--hostname=newhostname"},
+			curPrefs: &ipn.Prefs{
+				WantRunning:      true,
+				ControlURL:       ipn.DefaultControlURL,
+				RouteAll:         true,
+				AllowSingleHosts: false,
+				ExitNodeIP:       netaddr.MustParseIP("100.64.5.6"),
+				CorpDNS:          false,
+				ShieldsUp:        true,
+				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
+				Hostname:         "myhostname",
+				ForceDaemon:      true,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.0.0/16"),
+				},
+				NetfilterMode: preftype.NetfilterNoDivert,
+				OperatorUser:  "alice",
+			},
+			curUser: "eve",
+			want:    accidentalUpPrefix + " --hostname=newhostname --accept-dns=false --accept-routes --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --host-routes=false --netfilter-mode=nodivert --operator=alice --shields-up",
+		},
+		{
+			name:  "loggedout_is_implicit",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				LoggedOut:        true,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			want: "", // not an error. LoggedOut is implicit.
+		},
+		{
+			// Test that a pre-1.8 version of Tailscale with bogus NoSNAT pref
+			// values is able to enable exit nodes without warnings.
+			name:  "make_windows_exit_node",
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+
+				// And assume this no-op accidental pre-1.8 value:
+				NoSNAT: true,
+			},
+			goos: "windows",
+			want: "", // not an error
+		},
+		{
+			name:  "ignore_netfilter_change_non_linux",
+			flags: []string{"--accept-dns"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+
+				NetfilterMode: preftype.NetfilterNoDivert, // we never had this bug, but pretend it got set non-zero on Windows somehow
+			},
+			goos: "windows",
+			want: "", // not an error
+		},
+		{
+			name:  "operator_losing_routes_step1", // https://twitter.com/EXPbits/status/1390418145047887877
+			flags: []string{"--operator=expbits"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --operator=expbits --advertise-exit-node --advertise-routes=1.2.0.0/16",
+		},
+		{
+			name:  "operator_losing_routes_step2", // https://twitter.com/EXPbits/status/1390418145047887877
+			flags: []string{"--operator=expbits", "--advertise-routes=1.2.0.0/16"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-routes=1.2.0.0/16 --operator=expbits --advertise-exit-node",
+		},
+		{
+			name:  "errors_preserve_explicit_flags",
+			flags: []string{"--reset", "--force-reauth=false", "--authkey=secretrand"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      false,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+
+				Hostname: "foo",
+			},
+			want: accidentalUpPrefix + " --authkey=secretrand --force-reauth=false --reset --hostname=foo",
+		},
+		{
+			name:  "error_exit_node_omit_with_ip_pref",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				ExitNodeIP: netaddr.MustParseIP("100.64.5.4"),
+			},
+			want: accidentalUpPrefix + " --hostname=foo --exit-node=100.64.5.4",
+		},
+		{
+			name:          "error_exit_node_omit_with_id_pref",
+			flags:         []string{"--hostname=foo"},
+			curExitNodeIP: netaddr.MustParseIP("100.64.5.7"),
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				ExitNodeID: "some_stable_id",
+			},
+			want: accidentalUpPrefix + " --hostname=foo --exit-node=100.64.5.7",
+		},
+		{
+			name:  "ignore_login_server_synonym",
+			flags: []string{"--login-server=https://controlplane.tailscale.com"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       "https://login.tailscale.com",
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			want: "", // not an error
+		},
+		{
+			name:  "ignore_login_server_synonym_on_other_change",
+			flags: []string{"--netfilter-mode=off"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       "https://login.tailscale.com",
+				AllowSingleHosts: true,
+				CorpDNS:          false,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			want: accidentalUpPrefix + " --netfilter-mode=off --accept-dns=false",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			goos := "linux"
+			if tt.goos != "" {
+				goos = tt.goos
+			}
+			var upArgs upArgsT
+			flagSet := newUpFlagSet(goos, &upArgs)
+			flagSet.Parse(tt.flags)
+			newPrefs, err := prefsFromUpArgs(upArgs, t.Logf, new(ipnstate.Status), goos)
+			if err != nil {
+				t.Fatal(err)
+			}
+			applyImplicitPrefs(newPrefs, tt.curPrefs, tt.curUser)
+			var got string
+			if err := checkForAccidentalSettingReverts(newPrefs, tt.curPrefs, upCheckEnv{
+				goos:          goos,
+				flagSet:       flagSet,
+				curExitNodeIP: tt.curExitNodeIP,
+			}); err != nil {
+				got = err.Error()
+			}
+			if strings.TrimSpace(got) != tt.want {
+				t.Errorf("unexpected result\n got: %s\nwant: %s\n", got, tt.want)
+			}
+		})
+	}
+}
+
+func upArgsFromOSArgs(goos string, flagArgs ...string) (args upArgsT) {
+	fs := newUpFlagSet(goos, &args)
+	fs.Parse(flagArgs) // populates args
+	return
+}
+
+func TestPrefsFromUpArgs(t *testing.T) {
+	tests := []struct {
+		name     string
+		args     upArgsT
+		goos     string           // runtime.GOOS; empty means linux
+		st       *ipnstate.Status // or nil
+		want     *ipn.Prefs
+		wantErr  string
+		wantWarn string
+	}{
+		{
+			name: "default_linux",
+			goos: "linux",
+			args: upArgsFromOSArgs("linux"),
+			want: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      true,
+				NoSNAT:           false,
+				NetfilterMode:    preftype.NetfilterOn,
+				CorpDNS:          true,
+				AllowSingleHosts: true,
+			},
+		},
+		{
+			name: "default_windows",
+			goos: "windows",
+			args: upArgsFromOSArgs("windows"),
+			want: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      true,
+				CorpDNS:          true,
+				AllowSingleHosts: true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+		},
+		{
+			name: "advertise_default_route",
+			args: upArgsFromOSArgs("linux", "--advertise-exit-node"),
+			want: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      true,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+				NetfilterMode: preftype.NetfilterOn,
+			},
+		},
+		{
+			name: "error_advertise_route_invalid_ip",
+			args: upArgsT{
+				advertiseRoutes: "foo",
+			},
+			wantErr: `"foo" is not a valid IP address or CIDR prefix`,
+		},
+		{
+			name: "error_advertise_route_unmasked_bits",
+			args: upArgsT{
+				advertiseRoutes: "1.2.3.4/16",
+			},
+			wantErr: `1.2.3.4/16 has non-address bits set; expected 1.2.0.0/16`,
+		},
+		{
+			name: "error_exit_node_bad_ip",
+			args: upArgsT{
+				exitNodeIP: "foo",
+			},
+			wantErr: `invalid IP address "foo" for --exit-node: ParseIP("foo"): unable to parse IP`,
+		},
+		{
+			name: "error_exit_node_allow_lan_without_exit_node",
+			args: upArgsT{
+				exitNodeAllowLANAccess: true,
+			},
+			wantErr: `--exit-node-allow-lan-access can only be used with --exit-node`,
+		},
+		{
+			name: "error_tag_prefix",
+			args: upArgsT{
+				advertiseTags: "foo",
+			},
+			wantErr: `tag: "foo": tags must start with 'tag:'`,
+		},
+		{
+			name: "error_long_hostname",
+			args: upArgsT{
+				hostname: strings.Repeat("a", 300),
+			},
+			wantErr: `hostname too long: 300 bytes (max 256)`,
+		},
+		{
+			name: "error_linux_netfilter_empty",
+			args: upArgsT{
+				netfilterMode: "",
+			},
+			wantErr: `invalid value --netfilter-mode=""`,
+		},
+		{
+			name: "error_linux_netfilter_bogus",
+			args: upArgsT{
+				netfilterMode: "bogus",
+			},
+			wantErr: `invalid value --netfilter-mode="bogus"`,
+		},
+		{
+			name: "error_exit_node_ip_is_self_ip",
+			args: upArgsT{
+				exitNodeIP: "100.105.106.107",
+			},
+			st: &ipnstate.Status{
+				TailscaleIPs: []netaddr.IP{netaddr.MustParseIP("100.105.106.107")},
+			},
+			wantErr: `cannot use 100.105.106.107 as the exit node as it is a local IP address to this machine, did you mean --advertise-exit-node?`,
+		},
+		{
+			name: "warn_linux_netfilter_nodivert",
+			goos: "linux",
+			args: upArgsT{
+				netfilterMode: "nodivert",
+			},
+			wantWarn: "netfilter=nodivert; add iptables calls to ts-* chains manually.",
+			want: &ipn.Prefs{
+				WantRunning:   true,
+				NetfilterMode: preftype.NetfilterNoDivert,
+				NoSNAT:        true,
+			},
+		},
+		{
+			name: "warn_linux_netfilter_off",
+			goos: "linux",
+			args: upArgsT{
+				netfilterMode: "off",
+			},
+			wantWarn: "netfilter=off; configure iptables yourself.",
+			want: &ipn.Prefs{
+				WantRunning:   true,
+				NetfilterMode: preftype.NetfilterOff,
+				NoSNAT:        true,
+			},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var warnBuf bytes.Buffer
+			warnf := func(format string, a ...interface{}) {
+				fmt.Fprintf(&warnBuf, format, a...)
+			}
+			goos := tt.goos
+			if goos == "" {
+				goos = "linux"
+			}
+			st := tt.st
+			if st == nil {
+				st = new(ipnstate.Status)
+			}
+			got, err := prefsFromUpArgs(tt.args, warnf, st, goos)
+			gotErr := fmt.Sprint(err)
+			if tt.wantErr != "" {
+				if tt.wantErr != gotErr {
+					t.Errorf("wrong error.\n got error: %v\nwant error: %v\n", gotErr, tt.wantErr)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatal(err)
+			}
+			if tt.want == nil {
+				t.Fatal("tt.want is nil")
+			}
+			if !got.Equals(tt.want) {
+				jgot, _ := json.MarshalIndent(got, "", "\t")
+				jwant, _ := json.MarshalIndent(tt.want, "", "\t")
+				if bytes.Equal(jgot, jwant) {
+					t.Logf("prefs differ only in non-JSON-visible ways (nil/non-nil zero-length arrays)")
+				}
+				t.Errorf("wrong prefs\n got: %s\nwant: %s\n\ngot: %s\nwant: %s\n",
+					got.Pretty(), tt.want.Pretty(),
+					jgot, jwant,
+				)
+
+			}
+		})
+	}
+
+}
+
+func TestPrefFlagMapping(t *testing.T) {
+	prefHasFlag := map[string]bool{}
+	for _, pv := range prefsOfFlag {
+		for _, pref := range pv {
+			prefHasFlag[pref] = true
+		}
+	}
+
+	prefType := reflect.TypeOf(ipn.Prefs{})
+	for i := 0; i < prefType.NumField(); i++ {
+		prefName := prefType.Field(i).Name
+		if prefHasFlag[prefName] {
+			continue
+		}
+		switch prefName {
+		case "WantRunning", "Persist", "LoggedOut":
+			// All explicitly handled (ignored) by checkForAccidentalSettingReverts.
+			continue
+		case "OSVersion", "DeviceModel":
+			// Only used by Android, which doesn't have a CLI mode anyway, so
+			// fine to not map.
+			continue
+		case "NotepadURLs":
+			// TODO(bradfitz): https://github.com/tailscale/tailscale/issues/1830
+			continue
+		}
+		t.Errorf("unexpected new ipn.Pref field %q is not handled by up.go (see addPrefFlagMapping and checkForAccidentalSettingReverts)", prefName)
+	}
+}
+
+func TestFlagAppliesToOS(t *testing.T) {
+	for _, goos := range geese {
+		var upArgs upArgsT
+		fs := newUpFlagSet(goos, &upArgs)
+		fs.VisitAll(func(f *flag.Flag) {
+			if !flagAppliesToOS(f.Name, goos) {
+				t.Errorf("flagAppliesToOS(%q, %q) = false but found in %s set", f.Name, goos, goos)
+			}
+		})
+	}
+}
+
+func TestUpdatePrefs(t *testing.T) {
+	tests := []struct {
+		name     string
+		flags    []string // argv to be parsed into env.flagSet and env.upArgs
+		curPrefs *ipn.Prefs
+		env      upCheckEnv // empty goos means "linux"
+
+		wantSimpleUp   bool
+		wantJustEditMP *ipn.MaskedPrefs
+		wantErrSubtr   string
+	}{
+		{
+			name:  "bare_up_means_up",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL:  ipn.DefaultControlURL,
+				WantRunning: false,
+				Hostname:    "foo",
+			},
+		},
+		{
+			name:  "just_up",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL: ipn.DefaultControlURL,
+				Persist:    &persist.Persist{LoginName: "crawshaw.github"},
+			},
+			env: upCheckEnv{
+				backendState: "Stopped",
+			},
+			wantSimpleUp: true,
+		},
+		{
+			name:  "just_edit",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL: ipn.DefaultControlURL,
+				Persist:    &persist.Persist{LoginName: "crawshaw.github"},
+			},
+			env:            upCheckEnv{backendState: "Running"},
+			wantSimpleUp:   true,
+			wantJustEditMP: &ipn.MaskedPrefs{WantRunningSet: true},
+		},
+		{
+			name:  "control_synonym",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL: "https://login.tailscale.com",
+				Persist:    &persist.Persist{LoginName: "crawshaw.github"},
+			},
+			env:            upCheckEnv{backendState: "Running"},
+			wantSimpleUp:   true,
+			wantJustEditMP: &ipn.MaskedPrefs{WantRunningSet: true},
+		},
+		{
+			name:  "change_login_server",
+			flags: []string{"--login-server=https://localhost:1000"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       "https://login.tailscale.com",
+				Persist:          &persist.Persist{LoginName: "crawshaw.github"},
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			env:            upCheckEnv{backendState: "Running"},
+			wantSimpleUp:   true,
+			wantJustEditMP: &ipn.MaskedPrefs{WantRunningSet: true},
+			wantErrSubtr:   "can't change --login-server without --force-reauth",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if tt.env.goos == "" {
+				tt.env.goos = "linux"
+			}
+			tt.env.flagSet = newUpFlagSet(tt.env.goos, &tt.env.upArgs)
+			tt.env.flagSet.Parse(tt.flags)
+
+			newPrefs, err := prefsFromUpArgs(tt.env.upArgs, t.Logf, new(ipnstate.Status), tt.env.goos)
+			if err != nil {
+				t.Fatal(err)
+			}
+			simpleUp, justEditMP, err := updatePrefs(newPrefs, tt.curPrefs, tt.env)
+			if err != nil {
+				if tt.wantErrSubtr != "" {
+					if !strings.Contains(err.Error(), tt.wantErrSubtr) {
+						t.Fatalf("want error %q, got: %v", tt.wantErrSubtr, err)
+					}
+					return
+				}
+				t.Fatal(err)
+			}
+			if simpleUp != tt.wantSimpleUp {
+				t.Fatalf("simpleUp=%v, want %v", simpleUp, tt.wantSimpleUp)
+			}
+			if justEditMP != nil {
+				justEditMP.Prefs = ipn.Prefs{} // uninteresting
+			}
+			if !reflect.DeepEqual(justEditMP, tt.wantJustEditMP) {
+				t.Fatalf("justEditMP: %v", cmp.Diff(justEditMP, tt.wantJustEditMP))
+			}
+		})
+	}
+}

cmd/tailscale/cli/debug.go

@@ -13,11 +13,14 @@ import (
 	"io"
 	"log"
 	"os"
+	"runtime"
 	"strings"
 
 	"github.com/peterbourgon/ff/v2/ffcli"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/ipn"
+	"tailscale.com/paths"
+	"tailscale.com/safesocket"
 )
 
 var debugCmd = &ffcli.Command{
@@ -27,23 +30,57 @@ var debugCmd = &ffcli.Command{
 		fs := flag.NewFlagSet("debug", flag.ExitOnError)
 		fs.BoolVar(&debugArgs.goroutines, "daemon-goroutines", false, "If true, dump the tailscaled daemon's goroutines")
 		fs.BoolVar(&debugArgs.ipn, "ipn", false, "If true, subscribe to IPN notifications")
+		fs.BoolVar(&debugArgs.prefs, "prefs", false, "If true, dump active prefs")
+		fs.BoolVar(&debugArgs.derpMap, "derp", false, "If true, dump DERP map")
+		fs.BoolVar(&debugArgs.pretty, "pretty", false, "If true, pretty-print output (for --prefs)")
 		fs.BoolVar(&debugArgs.netMap, "netmap", true, "whether to include netmap in --ipn mode")
+		fs.BoolVar(&debugArgs.localCreds, "local-creds", false, "print how to connect to local tailscaled")
 		fs.StringVar(&debugArgs.file, "file", "", "get, delete:NAME, or NAME")
 		return fs
 	})(),
 }
 
 var debugArgs struct {
+	localCreds bool
 	goroutines bool
 	ipn        bool
 	netMap     bool
+	derpMap    bool
 	file       string
+	prefs      bool
+	pretty     bool
 }
 
 func runDebug(ctx context.Context, args []string) error {
 	if len(args) > 0 {
 		return errors.New("unknown arguments")
 	}
+	if debugArgs.localCreds {
+		port, token, err := safesocket.LocalTCPPortAndToken()
+		if err == nil {
+			fmt.Printf("curl -u:%s http://localhost:%d/localapi/v0/status\n", token, port)
+			return nil
+		}
+		if runtime.GOOS == "windows" {
+			fmt.Printf("curl http://localhost:41112/localapi/v0/status\n")
+			return nil
+		}
+		fmt.Printf("curl --unix-socket %s http://foo/localapi/v0/status\n", paths.DefaultTailscaledSocket())
+		return nil
+	}
+	if debugArgs.prefs {
+		prefs, err := tailscale.GetPrefs(ctx)
+		if err != nil {
+			return err
+		}
+		if debugArgs.pretty {
+			fmt.Println(prefs.Pretty())
+		} else {
+			j, _ := json.MarshalIndent(prefs, "", "\t")
+			fmt.Println(string(j))
+		}
+		return nil
+	}
 	if debugArgs.goroutines {
 		goroutines, err := tailscale.Goroutines(ctx)
 		if err != nil {
@@ -52,6 +89,18 @@ func runDebug(ctx context.Context, args []string) error {
 		os.Stdout.Write(goroutines)
 		return nil
 	}
+	if debugArgs.derpMap {
+		dm, err := tailscale.CurrentDERPMap(ctx)
+		if err != nil {
+			return fmt.Errorf(
+				"failed to get local derp map, instead `curl %s/derpmap/default`: %w", ipn.DefaultControlURL, err,
+			)
+		}
+		enc := json.NewEncoder(os.Stdout)
+		enc.SetIndent("", "\t")
+		enc.Encode(dm)
+		return nil
+	}
 	if debugArgs.ipn {
 		c, bc, ctx, cancel := connect(ctx)
 		defer cancel()

cmd/tailscale/cli/diag.go

@@ -0,0 +1,56 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build linux || windows || darwin
+// +build linux windows darwin
+
+package cli
+
+import (
+	"fmt"
+	"path/filepath"
+	"runtime"
+	"strings"
+
+	ps "github.com/mitchellh/go-ps"
+)
+
+// fixTailscaledConnectError is called when the local tailscaled has
+// been determined unreachable due to the provided origErr value. It
+// returns either the same error or a better one to help the user
+// understand why tailscaled isn't running for their platform.
+func fixTailscaledConnectError(origErr error) error {
+	procs, err := ps.Processes()
+	if err != nil {
+		return fmt.Errorf("failed to connect to local Tailscaled process and failed to enumerate processes while looking for it")
+	}
+	found := false
+	for _, proc := range procs {
+		base := filepath.Base(proc.Executable())
+		if base == "tailscaled" {
+			found = true
+			break
+		}
+		if runtime.GOOS == "darwin" && base == "IPNExtension" {
+			found = true
+			break
+		}
+		if runtime.GOOS == "windows" && strings.EqualFold(base, "tailscaled.exe") {
+			found = true
+			break
+		}
+	}
+	if !found {
+		switch runtime.GOOS {
+		case "windows":
+			return fmt.Errorf("failed to connect to local tailscaled process; is the Tailscale service running?")
+		case "darwin":
+			return fmt.Errorf("failed to connect to local Tailscale service; is Tailscale running?")
+		case "linux":
+			return fmt.Errorf("failed to connect to local tailscaled; it doesn't appear to be running (sudo systemctl start tailscaled ?)")
+		}
+		return fmt.Errorf("failed to connect to local tailscaled process; it doesn't appear to be running")
+	}
+	return fmt.Errorf("failed to connect to local tailscaled (which appears to be running). Got error: %w", origErr)
+}

cmd/tailscale/cli/diag_other.go

@@ -0,0 +1,17 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !linux && !windows && !darwin
+// +build !linux,!windows,!darwin
+
+package cli
+
+import "fmt"
+
+// The github.com/mitchellh/go-ps package doesn't work on all platforms,
+// so just don't diagnose connect failures.
+
+func fixTailscaledConnectError(origErr error) error {
+	return fmt.Errorf("failed to connect to local tailscaled process (is it running?); got: %w", origErr)
+}

cmd/tailscale/cli/down.go

@@ -8,7 +8,7 @@ import (
 	"context"
 	"fmt"
 	"log"
-	"time"
+	"os"
 
 	"github.com/peterbourgon/ff/v2/ffcli"
 	"tailscale.com/client/tailscale"
@@ -33,39 +33,14 @@ func runDown(ctx context.Context, args []string) error {
 		return fmt.Errorf("error fetching current status: %w", err)
 	}
 	if st.BackendState == "Stopped" {
-		log.Printf("already stopped")
+		fmt.Fprintf(os.Stderr, "Tailscale was already stopped.\n")
 		return nil
 	}
-	log.Printf("was in state %q", st.BackendState)
-
-	c, bc, ctx, cancel := connect(ctx)
-	defer cancel()
-
-	timer := time.AfterFunc(5*time.Second, func() {
-		log.Fatalf("timeout running stop")
-	})
-	defer timer.Stop()
-
-	bc.SetNotifyCallback(func(n ipn.Notify) {
-		if n.ErrMessage != nil {
-			log.Fatal(*n.ErrMessage)
-		}
-		if n.State != nil {
-			log.Printf("now in state %q", *n.State)
-			if *n.State == ipn.Stopped {
-				cancel()
-			}
-			return
-		}
-	})
-
-	bc.EditPrefs(&ipn.MaskedPrefs{
+	_, err = tailscale.EditPrefs(ctx, &ipn.MaskedPrefs{
 		Prefs: ipn.Prefs{
 			WantRunning: false,
 		},
 		WantRunningSet: true,
 	})
-	pump(ctx, bc, c)
-
-	return nil
+	return err
 }

cmd/tailscale/cli/file.go

@@ -0,0 +1,436 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"log"
+	"mime"
+	"net/http"
+	"net/url"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"time"
+	"unicode/utf8"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"golang.org/x/time/rate"
+	"inet.af/netaddr"
+	"tailscale.com/client/tailscale"
+	"tailscale.com/client/tailscale/apitype"
+	"tailscale.com/ipn"
+	"tailscale.com/net/tsaddr"
+	"tailscale.com/version"
+)
+
+var fileCmd = &ffcli.Command{
+	Name:       "file",
+	ShortUsage: "file <cp|get> ...",
+	ShortHelp:  "Send or receive files",
+	Subcommands: []*ffcli.Command{
+		fileCpCmd,
+		fileGetCmd,
+	},
+	Exec: func(context.Context, []string) error {
+		// TODO(bradfitz): is there a better ffcli way to
+		// annotate subcommand-required commands that don't
+		// have an exec body of their own?
+		return errors.New("file subcommand required; run 'tailscale file -h' for details")
+	},
+}
+
+var fileCpCmd = &ffcli.Command{
+	Name:       "cp",
+	ShortUsage: "file cp <files...> <target>:",
+	ShortHelp:  "Copy file(s) to a host",
+	Exec:       runCp,
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("cp", flag.ExitOnError)
+		fs.StringVar(&cpArgs.name, "name", "", "alternate filename to use, especially useful when <file> is \"-\" (stdin)")
+		fs.BoolVar(&cpArgs.verbose, "verbose", false, "verbose output")
+		fs.BoolVar(&cpArgs.targets, "targets", false, "list possible file cp targets")
+		return fs
+	})(),
+}
+
+var cpArgs struct {
+	name    string
+	verbose bool
+	targets bool
+}
+
+func runCp(ctx context.Context, args []string) error {
+	if cpArgs.targets {
+		return runCpTargets(ctx, args)
+	}
+	if len(args) < 2 {
+		return errors.New("usage: tailscale file cp <files...> <target>:")
+	}
+	files, target := args[:len(args)-1], args[len(args)-1]
+	if !strings.HasSuffix(target, ":") {
+		return fmt.Errorf("final argument to 'tailscale file cp' must end in colon")
+	}
+	target = strings.TrimSuffix(target, ":")
+	hadBrackets := false
+	if strings.HasPrefix(target, "[") && strings.HasSuffix(target, "]") {
+		hadBrackets = true
+		target = strings.TrimSuffix(strings.TrimPrefix(target, "["), "]")
+	}
+	if ip, err := netaddr.ParseIP(target); err == nil && ip.Is6() && !hadBrackets {
+		return fmt.Errorf("an IPv6 literal must be written as [%s]", ip)
+	} else if hadBrackets && (err != nil || !ip.Is6()) {
+		return errors.New("unexpected brackets around target")
+	}
+	ip, err := tailscaleIPFromArg(ctx, target)
+	if err != nil {
+		return err
+	}
+
+	peerAPIBase, isOffline, err := discoverPeerAPIBase(ctx, ip)
+	if err != nil {
+		return fmt.Errorf("can't send to %s: %v", target, err)
+	}
+	if isOffline {
+		fmt.Fprintf(os.Stderr, "# warning: %s is offline\n", target)
+	}
+
+	if len(files) > 1 {
+		if cpArgs.name != "" {
+			return errors.New("can't use --name= with multiple files")
+		}
+		for _, fileArg := range files {
+			if fileArg == "-" {
+				return errors.New("can't use '-' as STDIN file when providing filename arguments")
+			}
+		}
+	}
+
+	for _, fileArg := range files {
+		var fileContents io.Reader
+		var name = cpArgs.name
+		var contentLength int64 = -1
+		if fileArg == "-" {
+			fileContents = os.Stdin
+			if name == "" {
+				name, fileContents, err = pickStdinFilename()
+				if err != nil {
+					return err
+				}
+			}
+		} else {
+			f, err := os.Open(fileArg)
+			if err != nil {
+				if version.IsSandboxedMacOS() {
+					return errors.New("the GUI version of Tailscale on macOS runs in a macOS sandbox that can't read files")
+				}
+				return err
+			}
+			defer f.Close()
+			fi, err := f.Stat()
+			if err != nil {
+				return err
+			}
+			if fi.IsDir() {
+				return errors.New("directories not supported")
+			}
+			contentLength = fi.Size()
+			fileContents = io.LimitReader(f, contentLength)
+			if name == "" {
+				name = filepath.Base(fileArg)
+			}
+
+			if slow, _ := strconv.ParseBool(os.Getenv("TS_DEBUG_SLOW_PUSH")); slow {
+				fileContents = &slowReader{r: fileContents}
+			}
+		}
+
+		dstURL := peerAPIBase + "/v0/put/" + url.PathEscape(name)
+		req, err := http.NewRequestWithContext(ctx, "PUT", dstURL, fileContents)
+		if err != nil {
+			return err
+		}
+		req.ContentLength = contentLength
+		if cpArgs.verbose {
+			log.Printf("sending to %v ...", dstURL)
+		}
+		res, err := http.DefaultClient.Do(req)
+		if err != nil {
+			return err
+		}
+		if res.StatusCode == 200 {
+			io.Copy(ioutil.Discard, res.Body)
+			res.Body.Close()
+			continue
+		}
+		io.Copy(os.Stdout, res.Body)
+		res.Body.Close()
+		return errors.New(res.Status)
+	}
+	return nil
+}
+
+func discoverPeerAPIBase(ctx context.Context, ipStr string) (base string, isOffline bool, err error) {
+	ip, err := netaddr.ParseIP(ipStr)
+	if err != nil {
+		return "", false, err
+	}
+	fts, err := tailscale.FileTargets(ctx)
+	if err != nil {
+		return "", false, err
+	}
+	for _, ft := range fts {
+		n := ft.Node
+		for _, a := range n.Addresses {
+			if a.IP() != ip {
+				continue
+			}
+			isOffline = n.Online != nil && !*n.Online
+			return ft.PeerAPIURL, isOffline, nil
+		}
+	}
+	return "", false, fileTargetErrorDetail(ctx, ip)
+}
+
+// fileTargetErrorDetail returns a non-nil error saying why ip is an
+// invalid file sharing target.
+func fileTargetErrorDetail(ctx context.Context, ip netaddr.IP) error {
+	found := false
+	if st, err := tailscale.Status(ctx); err == nil && st.Self != nil {
+		for _, peer := range st.Peer {
+			for _, pip := range peer.TailscaleIPs {
+				if pip == ip {
+					found = true
+					if peer.UserID != st.Self.UserID {
+						return errors.New("owned by different user; can only send files to your own devices")
+					}
+				}
+			}
+		}
+	}
+	if found {
+		return errors.New("target seems to be running an old Tailscale version")
+	}
+	if !tsaddr.IsTailscaleIP(ip) {
+		return fmt.Errorf("unknown target; %v is not a Tailscale IP address", ip)
+	}
+	return errors.New("unknown target; not in your Tailnet")
+}
+
+const maxSniff = 4 << 20
+
+func ext(b []byte) string {
+	if len(b) < maxSniff && utf8.Valid(b) {
+		return ".txt"
+	}
+	if exts, _ := mime.ExtensionsByType(http.DetectContentType(b)); len(exts) > 0 {
+		return exts[0]
+	}
+	return ""
+}
+
+// pickStdinFilename reads a bit of stdin to return a good filename
+// for its contents. The returned Reader is the concatenation of the
+// read and unread bits.
+func pickStdinFilename() (name string, r io.Reader, err error) {
+	sniff, err := io.ReadAll(io.LimitReader(os.Stdin, maxSniff))
+	if err != nil {
+		return "", nil, err
+	}
+	return "stdin" + ext(sniff), io.MultiReader(bytes.NewReader(sniff), os.Stdin), nil
+}
+
+type slowReader struct {
+	r  io.Reader
+	rl *rate.Limiter
+}
+
+func (r *slowReader) Read(p []byte) (n int, err error) {
+	const burst = 4 << 10
+	plen := len(p)
+	if plen > burst {
+		plen = burst
+	}
+	if r.rl == nil {
+		r.rl = rate.NewLimiter(rate.Limit(1<<10), burst)
+	}
+	n, err = r.r.Read(p[:plen])
+	r.rl.WaitN(context.Background(), n)
+	return
+}
+
+func runCpTargets(ctx context.Context, args []string) error {
+	if len(args) > 0 {
+		return errors.New("invalid arguments with --targets")
+	}
+	fts, err := tailscale.FileTargets(ctx)
+	if err != nil {
+		return err
+	}
+	for _, ft := range fts {
+		n := ft.Node
+		var detail string
+		if n.Online != nil {
+			if !*n.Online {
+				detail = "offline"
+			}
+		} else {
+			detail = "unknown-status"
+		}
+		if detail != "" && n.LastSeen != nil {
+			d := time.Since(*n.LastSeen)
+			detail += fmt.Sprintf("; last seen %v ago", d.Round(time.Minute))
+		}
+		if detail != "" {
+			detail = "\t" + detail
+		}
+		fmt.Printf("%s\t%s%s\n", n.Addresses[0].IP(), n.ComputedName, detail)
+	}
+	return nil
+}
+
+var fileGetCmd = &ffcli.Command{
+	Name:       "get",
+	ShortUsage: "file get [--wait] [--verbose] <target-directory>",
+	ShortHelp:  "Move files out of the Tailscale file inbox",
+	Exec:       runFileGet,
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("get", flag.ExitOnError)
+		fs.BoolVar(&getArgs.wait, "wait", false, "wait for a file to arrive if inbox is empty")
+		fs.BoolVar(&getArgs.verbose, "verbose", false, "verbose output")
+		return fs
+	})(),
+}
+
+var getArgs struct {
+	wait    bool
+	verbose bool
+}
+
+func runFileGet(ctx context.Context, args []string) error {
+	if len(args) != 1 {
+		return errors.New("usage: file get <target-directory>")
+	}
+	log.SetFlags(0)
+
+	dir := args[0]
+	if dir == "/dev/null" {
+		return wipeInbox(ctx)
+	}
+
+	if fi, err := os.Stat(dir); err != nil || !fi.IsDir() {
+		return fmt.Errorf("%q is not a directory", dir)
+	}
+
+	var wfs []apitype.WaitingFile
+	var err error
+	for {
+		wfs, err = tailscale.WaitingFiles(ctx)
+		if err != nil {
+			return fmt.Errorf("getting WaitingFiles: %v", err)
+		}
+		if len(wfs) != 0 || !getArgs.wait {
+			break
+		}
+		if getArgs.verbose {
+			log.Printf("waiting for file...")
+		}
+		if err := waitForFile(ctx); err != nil {
+			return err
+		}
+	}
+
+	deleted := 0
+	for _, wf := range wfs {
+		rc, size, err := tailscale.GetWaitingFile(ctx, wf.Name)
+		if err != nil {
+			return fmt.Errorf("opening inbox file %q: %v", wf.Name, err)
+		}
+		targetFile := filepath.Join(dir, wf.Name)
+		of, err := os.OpenFile(targetFile, os.O_RDWR|os.O_CREATE|os.O_EXCL, 0644)
+		if err != nil {
+			if _, err := os.Stat(targetFile); err == nil {
+				return fmt.Errorf("refusing to overwrite %v", targetFile)
+			}
+			return err
+		}
+		_, err = io.Copy(of, rc)
+		rc.Close()
+		if err != nil {
+			return fmt.Errorf("failed to write %v: %v", targetFile, err)
+		}
+		if err := of.Close(); err != nil {
+			return err
+		}
+		if getArgs.verbose {
+			log.Printf("wrote %v (%d bytes)", wf.Name, size)
+		}
+		if err := tailscale.DeleteWaitingFile(ctx, wf.Name); err != nil {
+			return fmt.Errorf("deleting %q from inbox: %v", wf.Name, err)
+		}
+		deleted++
+	}
+	if getArgs.verbose {
+		log.Printf("moved %d files", deleted)
+	}
+	return nil
+}
+
+func wipeInbox(ctx context.Context) error {
+	if getArgs.wait {
+		return errors.New("can't use --wait with /dev/null target")
+	}
+	wfs, err := tailscale.WaitingFiles(ctx)
+	if err != nil {
+		return fmt.Errorf("getting WaitingFiles: %v", err)
+	}
+	deleted := 0
+	for _, wf := range wfs {
+		if getArgs.verbose {
+			log.Printf("deleting %v ...", wf.Name)
+		}
+		if err := tailscale.DeleteWaitingFile(ctx, wf.Name); err != nil {
+			return fmt.Errorf("deleting %q: %v", wf.Name, err)
+		}
+		deleted++
+	}
+	if getArgs.verbose {
+		log.Printf("deleted %d files", deleted)
+	}
+	return nil
+}
+
+func waitForFile(ctx context.Context) error {
+	c, bc, pumpCtx, cancel := connect(ctx)
+	defer cancel()
+	fileWaiting := make(chan bool, 1)
+	bc.SetNotifyCallback(func(n ipn.Notify) {
+		if n.ErrMessage != nil {
+			log.Fatal(*n.ErrMessage)
+		}
+		if n.FilesWaiting != nil {
+			select {
+			case fileWaiting <- true:
+			default:
+			}
+		}
+	})
+	go pump(pumpCtx, bc, c)
+	select {
+	case <-fileWaiting:
+		return nil
+	case <-pumpCtx.Done():
+		return pumpCtx.Err()
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}

cmd/tailscale/cli/ip.go

@@ -11,13 +11,16 @@ import (
 	"fmt"
 
 	"github.com/peterbourgon/ff/v2/ffcli"
+	"inet.af/netaddr"
 	"tailscale.com/client/tailscale"
+	"tailscale.com/ipn/ipnstate"
 )
 
 var ipCmd = &ffcli.Command{
 	Name:       "ip",
-	ShortUsage: "ip [-4] [-6]",
-	ShortHelp:  "Show this machine's current Tailscale IP address(es)",
+	ShortUsage: "ip [-4] [-6] [peername]",
+	ShortHelp:  "Show current Tailscale IP address(es)",
+	LongHelp:   "Shows the Tailscale IP address of the current machine without an argument. With an argument, it shows the IP of a named peer.",
 	Exec:       runIP,
 	FlagSet: (func() *flag.FlagSet {
 		fs := flag.NewFlagSet("ip", flag.ExitOnError)
@@ -33,9 +36,14 @@ var ipArgs struct {
 }
 
 func runIP(ctx context.Context, args []string) error {
-	if len(args) > 0 {
+	if len(args) > 1 {
 		return errors.New("unknown arguments")
 	}
+	var of string
+	if len(args) == 1 {
+		of = args[0]
+	}
+
 	v4, v6 := ipArgs.want4, ipArgs.want6
 	if v4 && v6 {
 		return errors.New("tailscale up -4 and -6 are mutually exclusive")
@@ -47,11 +55,24 @@ func runIP(ctx context.Context, args []string) error {
 	if err != nil {
 		return err
 	}
-	if len(st.TailscaleIPs) == 0 {
+	ips := st.TailscaleIPs
+	if of != "" {
+		ip, err := tailscaleIPFromArg(ctx, of)
+		if err != nil {
+			return err
+		}
+		peer, ok := peerMatchingIP(st, ip)
+		if !ok {
+			return fmt.Errorf("no peer found with IP %v", ip)
+		}
+		ips = peer.TailscaleIPs
+	}
+	if len(ips) == 0 {
 		return fmt.Errorf("no current Tailscale IPs; state: %v", st.BackendState)
 	}
+
 	match := false
-	for _, ip := range st.TailscaleIPs {
+	for _, ip := range ips {
 		if ip.Is4() && v4 || ip.Is6() && v6 {
 			match = true
 			fmt.Println(ip)
@@ -67,3 +88,18 @@ func runIP(ctx context.Context, args []string) error {
 	}
 	return nil
 }
+
+func peerMatchingIP(st *ipnstate.Status, ipStr string) (ps *ipnstate.PeerStatus, ok bool) {
+	ip, err := netaddr.ParseIP(ipStr)
+	if err != nil {
+		return
+	}
+	for _, ps = range st.Peer {
+		for _, pip := range ps.TailscaleIPs {
+			if ip == pip {
+				return ps, true
+			}
+		}
+	}
+	return nil, false
+}

cmd/tailscale/cli/logout.go

@@ -0,0 +1,34 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"context"
+	"log"
+	"strings"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"tailscale.com/client/tailscale"
+)
+
+var logoutCmd = &ffcli.Command{
+	Name:       "logout",
+	ShortUsage: "logout [flags]",
+	ShortHelp:  "Disconnect from Tailscale and expire current node key",
+
+	LongHelp: strings.TrimSpace(`
+"tailscale logout" brings the network down and invalidates
+the current node key, forcing a future use of it to cause
+a reauthentication.
+`),
+	Exec: runLogout,
+}
+
+func runLogout(ctx context.Context, args []string) error {
+	if len(args) > 0 {
+		log.Fatalf("too many non-flag arguments: %q", args)
+	}
+	return tailscale.Logout(ctx)
+}

cmd/tailscale/cli/netcheck.go

@@ -9,14 +9,18 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"io"
+	"io/ioutil"
 	"log"
+	"net/http"
 	"os"
 	"sort"
 	"strings"
 	"time"
 
 	"github.com/peterbourgon/ff/v2/ffcli"
-	"tailscale.com/derp/derpmap"
+	"tailscale.com/client/tailscale"
+	"tailscale.com/ipn"
 	"tailscale.com/net/netcheck"
 	"tailscale.com/net/portmapper"
 	"tailscale.com/tailcfg"
@@ -46,7 +50,7 @@ var netcheckArgs struct {
 func runNetcheck(ctx context.Context, args []string) error {
 	c := &netcheck.Client{
 		UDPBindAddr: os.Getenv("TS_DEBUG_NETCHECK_UDP_BIND"),
-		PortMapper:  portmapper.NewClient(logger.WithPrefix(log.Printf, "portmap: ")),
+		PortMapper:  portmapper.NewClient(logger.WithPrefix(log.Printf, "portmap: "), nil),
 	}
 	if netcheckArgs.verbose {
 		c.Logf = logger.WithPrefix(log.Printf, "netcheck: ")
@@ -59,7 +63,13 @@ func runNetcheck(ctx context.Context, args []string) error {
 		fmt.Fprintln(os.Stderr, "# Warning: this JSON format is not yet considered a stable interface")
 	}
 
-	dm := derpmap.Prod()
+	dm, err := tailscale.CurrentDERPMap(ctx)
+	if err != nil {
+		dm, err = prodDERPMap(ctx, http.DefaultClient)
+		if err != nil {
+			return err
+		}
+	}
 	for {
 		t0 := time.Now()
 		report, err := c.GetReport(ctx, dm)
@@ -176,3 +186,27 @@ func portMapping(r *netcheck.Report) string {
 	}
 	return strings.Join(got, ", ")
 }
+
+func prodDERPMap(ctx context.Context, httpc *http.Client) (*tailcfg.DERPMap, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", ipn.DefaultControlURL+"/derpmap/default", nil)
+	if err != nil {
+		return nil, fmt.Errorf("create prodDERPMap request: %w", err)
+	}
+	res, err := httpc.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("fetch prodDERPMap failed: %w", err)
+	}
+	defer res.Body.Close()
+	b, err := ioutil.ReadAll(io.LimitReader(res.Body, 1<<20))
+	if err != nil {
+		return nil, fmt.Errorf("fetch prodDERPMap failed: %w", err)
+	}
+	if res.StatusCode != 200 {
+		return nil, fmt.Errorf("fetch prodDERPMap: %v: %s", res.Status, b)
+	}
+	var derpMap tailcfg.DERPMap
+	if err = json.Unmarshal(b, &derpMap); err != nil {
+		return nil, fmt.Errorf("fetch prodDERPMap: %w", err)
+	}
+	return &derpMap, nil
+}

cmd/tailscale/cli/ping.go

@@ -80,7 +80,8 @@ func runPing(ctx context.Context, args []string) error {
 			prc <- pr
 		}
 	})
-	go pump(ctx, bc, c)
+	pumpErr := make(chan error, 1)
+	go func() { pumpErr <- pump(ctx, bc, c) }()
 
 	hostOrIP := args[0]
 	ip, err := tailscaleIPFromArg(ctx, hostOrIP)
@@ -101,6 +102,8 @@ func runPing(ctx context.Context, args []string) error {
 		select {
 		case <-timer.C:
 			fmt.Printf("timeout waiting for ping reply\n")
+		case err := <-pumpErr:
+			return err
 		case pr := <-prc:
 			timer.Stop()
 			if pr.Err != "" {
@@ -136,6 +139,9 @@ func runPing(ctx context.Context, args []string) error {
 			if !anyPong {
 				return errors.New("no reply")
 			}
+			if pingArgs.untilDirect {
+				return errors.New("direct connection not established")
+			}
 			return nil
 		}
 	}
@@ -154,7 +160,10 @@ func tailscaleIPFromArg(ctx context.Context, hostOrIP string) (ip string, err er
 	}
 	for _, ps := range st.Peer {
 		if hostOrIP == dnsOrQuoteHostname(st, ps) || hostOrIP == ps.DNSName {
-			return ps.TailAddr, nil
+			if len(ps.TailscaleIPs) == 0 {
+				return "", errors.New("node found but lacks an IP")
+			}
+			return ps.TailscaleIPs[0].String(), nil
 		}
 	}
 

cmd/tailscale/cli/push.go

@@ -1,173 +0,0 @@
-// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package cli
-
-import (
-	"bytes"
-	"context"
-	"errors"
-	"flag"
-	"fmt"
-	"io"
-	"log"
-	"mime"
-	"net"
-	"net/http"
-	"net/url"
-	"os"
-	"time"
-	"unicode/utf8"
-
-	"github.com/peterbourgon/ff/v2/ffcli"
-	"tailscale.com/ipn"
-	"tailscale.com/ipn/ipnstate"
-)
-
-var pushCmd = &ffcli.Command{
-	Name:       "push",
-	ShortUsage: "push [--flags] <hostname-or-IP> <file>",
-	ShortHelp:  "Push a file to a host",
-	Exec:       runPush,
-	FlagSet: (func() *flag.FlagSet {
-		fs := flag.NewFlagSet("push", flag.ExitOnError)
-		fs.StringVar(&pushArgs.name, "name", "", "alternate filename to use, especially useful when <file> is \"-\" (stdin)")
-		fs.BoolVar(&pushArgs.verbose, "verbose", false, "verbose output")
-		return fs
-	})(),
-}
-
-var pushArgs struct {
-	name    string
-	verbose bool
-}
-
-func runPush(ctx context.Context, args []string) error {
-	if len(args) != 2 || args[0] == "" {
-		return errors.New("usage: push <hostname-or-IP> <file>")
-	}
-	var ip string
-
-	hostOrIP, fileArg := args[0], args[1]
-	ip, err := tailscaleIPFromArg(ctx, hostOrIP)
-	if err != nil {
-		return err
-	}
-
-	peerAPIPort, err := discoverPeerAPIPort(ctx, ip)
-	if err != nil {
-		return err
-	}
-
-	var fileContents io.Reader
-	var name = pushArgs.name
-	if fileArg == "-" {
-		fileContents = os.Stdin
-		if name == "" {
-			name, fileContents, err = pickStdinFilename()
-			if err != nil {
-				return err
-			}
-		}
-	} else {
-		f, err := os.Open(fileArg)
-		if err != nil {
-			return err
-		}
-		defer f.Close()
-		fileContents = f
-		if name == "" {
-			name = fileArg
-		}
-	}
-
-	dstURL := "http://" + net.JoinHostPort(ip, fmt.Sprint(peerAPIPort)) + "/v0/put/" + url.PathEscape(name)
-	req, err := http.NewRequestWithContext(ctx, "PUT", dstURL, fileContents)
-	if err != nil {
-		return err
-	}
-	if pushArgs.verbose {
-		log.Printf("sending to %v ...", dstURL)
-	}
-	res, err := http.DefaultClient.Do(req)
-	if err != nil {
-		return err
-	}
-	defer res.Body.Close()
-	if res.StatusCode == 200 {
-		return nil
-	}
-	io.Copy(os.Stdout, res.Body)
-	return errors.New(res.Status)
-}
-
-func discoverPeerAPIPort(ctx context.Context, ip string) (port uint16, err error) {
-	c, bc, ctx, cancel := connect(ctx)
-	defer cancel()
-
-	prc := make(chan *ipnstate.PingResult, 2)
-	bc.SetNotifyCallback(func(n ipn.Notify) {
-		if n.ErrMessage != nil {
-			log.Fatal(*n.ErrMessage)
-		}
-		if pr := n.PingResult; pr != nil && pr.IP == ip {
-			prc <- pr
-		}
-	})
-	go pump(ctx, bc, c)
-
-	ticker := time.NewTicker(time.Second)
-	defer ticker.Stop()
-
-	discoPings := 0
-	timer := time.NewTimer(10 * time.Second)
-	defer timer.Stop()
-
-	sendPings := func() {
-		bc.Ping(ip, false)
-		bc.Ping(ip, true)
-	}
-	sendPings()
-	for {
-		select {
-		case <-ticker.C:
-			sendPings()
-		case <-timer.C:
-			return 0, fmt.Errorf("timeout contacting %v; it offline?", ip)
-		case pr := <-prc:
-			if p := pr.PeerAPIPort; p != 0 {
-				return p, nil
-			}
-			discoPings++
-			if discoPings == 3 {
-				return 0, fmt.Errorf("%v is online, but seems to be running an old Tailscale version", ip)
-			}
-		case <-ctx.Done():
-			return 0, ctx.Err()
-		}
-	}
-}
-
-const maxSniff = 4 << 20
-
-func ext(b []byte) string {
-	if len(b) < maxSniff && utf8.Valid(b) {
-		return ".txt"
-	}
-	if exts, _ := mime.ExtensionsByType(http.DetectContentType(b)); len(exts) > 0 {
-		return exts[0]
-	}
-	return ""
-}
-
-// pickStdinFilename reads a bit of stdin to return a good filename
-// for its contents. The returned Reader is the concatenation of the
-// read and unread bits.
-func pickStdinFilename() (name string, r io.Reader, err error) {
-	sniff, err := io.ReadAll(io.LimitReader(os.Stdin, maxSniff))
-	if err != nil {
-		return "", nil, err
-	}
-	return "stdin" + ext(sniff), io.MultiReader(bytes.NewReader(sniff), os.Stdin), nil
-}

cmd/tailscale/cli/status.go

@@ -14,10 +14,10 @@ import (
 	"net/http"
 	"os"
 	"strings"
-	"time"
 
 	"github.com/peterbourgon/ff/v2/ffcli"
 	"github.com/toqueteos/webbrowser"
+	"inet.af/netaddr"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
@@ -56,12 +56,12 @@ var statusArgs struct {
 func runStatus(ctx context.Context, args []string) error {
 	st, err := tailscale.Status(ctx)
 	if err != nil {
-		return err
+		return fixTailscaledConnectError(err)
 	}
 	if statusArgs.json {
 		if statusArgs.active {
 			for peer, ps := range st.Peer {
-				if !peerActive(ps) {
+				if !ps.Active {
 					delete(st.Peer, peer)
 				}
 			}
@@ -106,24 +106,38 @@ func runStatus(ctx context.Context, args []string) error {
 		return err
 	}
 
-	if st.BackendState == ipn.Stopped.String() {
+	switch st.BackendState {
+	default:
+		fmt.Fprintf(os.Stderr, "unexpected state: %s\n", st.BackendState)
+		os.Exit(1)
+	case ipn.Stopped.String():
 		fmt.Println("Tailscale is stopped.")
 		os.Exit(1)
+	case ipn.NeedsLogin.String():
+		fmt.Println("Logged out.")
+		if st.AuthURL != "" {
+			fmt.Printf("\nLog in at: %s\n", st.AuthURL)
+		}
+		os.Exit(1)
+	case ipn.NeedsMachineAuth.String():
+		fmt.Println("Machine is not yet authorized by tailnet admin.")
+		os.Exit(1)
+	case ipn.Running.String():
+		// Run below.
 	}
 
 	var buf bytes.Buffer
 	f := func(format string, a ...interface{}) { fmt.Fprintf(&buf, format, a...) }
 	printPS := func(ps *ipnstate.PeerStatus) {
-		active := peerActive(ps)
 		f("%-15s %-20s %-12s %-7s ",
-			ps.TailAddr,
+			firstIPString(ps.TailscaleIPs),
 			dnsOrQuoteHostname(st, ps),
 			ownerLogin(st, ps),
 			ps.OS,
 		)
 		relay := ps.Relay
 		anyTraffic := ps.TxBytes != 0 || ps.RxBytes != 0
-		if !active {
+		if !ps.Active {
 			if ps.ExitNode {
 				f("idle; exit node")
 			} else if anyTraffic {
@@ -162,8 +176,7 @@ func runStatus(ctx context.Context, args []string) error {
 		}
 		ipnstate.SortPeers(peers)
 		for _, ps := range peers {
-			active := peerActive(ps)
-			if statusArgs.active && !active {
+			if statusArgs.active && !ps.Active {
 				continue
 			}
 			printPS(ps)
@@ -173,13 +186,6 @@ func runStatus(ctx context.Context, args []string) error {
 	return nil
 }
 
-// peerActive reports whether ps has recent activity.
-//
-// TODO: have the server report this bool instead.
-func peerActive(ps *ipnstate.PeerStatus) bool {
-	return !ps.LastWrite.IsZero() && time.Since(ps.LastWrite) < 2*time.Minute
-}
-
 func dnsOrQuoteHostname(st *ipnstate.Status, ps *ipnstate.PeerStatus) string {
 	baseName := dnsname.TrimSuffix(ps.DNSName, st.MagicDNSSuffix)
 	if baseName != "" {
@@ -201,3 +207,10 @@ func ownerLogin(st *ipnstate.Status, ps *ipnstate.PeerStatus) string {
 	}
 	return u.LoginName
 }
+
+func firstIPString(v []netaddr.IP) string {
+	if len(v) == 0 {
+		return ""
+	}
+	return v[0].String()
+}

cmd/tailscale/cli/web.go

@@ -9,12 +9,16 @@ import (
 	"context"
 	_ "embed"
 	"encoding/json"
+	"encoding/xml"
 	"flag"
 	"fmt"
 	"html/template"
+	"io/ioutil"
 	"log"
+	"net"
 	"net/http"
 	"net/http/cgi"
+	"net/url"
 	"os/exec"
 	"runtime"
 	"strings"
@@ -24,6 +28,7 @@ import (
 	"tailscale.com/ipn"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/preftype"
+	"tailscale.com/util/groupmember"
 	"tailscale.com/version/distro"
 )
 
@@ -33,6 +38,9 @@ var webHTML string
 //go:embed web.css
 var webCSS string
 
+//go:embed auth-redirect.html
+var authenticationRedirectHTML string
+
 var tmpl *template.Template
 
 func init() {
@@ -53,6 +61,14 @@ var webCmd = &ffcli.Command{
 	ShortUsage: "web [flags]",
 	ShortHelp:  "Run a web server for controlling Tailscale",
 
+	LongHelp: strings.TrimSpace(`
+"tailscale web" runs a webserver for controlling the Tailscale daemon.
+
+It's primarily intended for use on Synology, QNAP, and other
+NAS devices where a web interface is the natural place to control
+Tailscale, as opposed to a CLI or a native app.
+`),
+
 	FlagSet: (func() *flag.FlagSet {
 		webf := flag.NewFlagSet("web", flag.ExitOnError)
 		webf.StringVar(&webArgs.listen, "listen", "localhost:8088", "listen address; use port 0 for automatic")
@@ -73,28 +89,134 @@ func runWeb(ctx context.Context, args []string) error {
 	}
 
 	if webArgs.cgi {
-		return cgi.Serve(http.HandlerFunc(webHandler))
+		if err := cgi.Serve(http.HandlerFunc(webHandler)); err != nil {
+			log.Printf("tailscale.cgi: %v", err)
+			return err
+		}
+		return nil
 	}
+
+	log.Printf("web server running on: %s", urlOfListenAddr(webArgs.listen))
 	return http.ListenAndServe(webArgs.listen, http.HandlerFunc(webHandler))
 }
 
-func auth() (string, error) {
-	if distro.Get() == distro.Synology {
-		cmd := exec.Command("/usr/syno/synoman/webman/modules/authenticate.cgi")
-		out, err := cmd.CombinedOutput()
-		if err != nil {
-			return "", fmt.Errorf("auth: %v: %s", err, out)
-		}
-		return string(out), nil
+// urlOfListenAddr parses a given listen address into a formatted URL
+func urlOfListenAddr(addr string) string {
+	host, port, _ := net.SplitHostPort(addr)
+	if host == "" {
+		host = "127.0.0.1"
 	}
+	return fmt.Sprintf("http://%s", net.JoinHostPort(host, port))
+}
 
+// authorize returns the name of the user accessing the web UI after verifying
+// whether the user has access to the web UI. The function will write the
+// error to the provided http.ResponseWriter.
+// Note: This is different from a tailscale user, and is typically the local
+// user on the node.
+func authorize(w http.ResponseWriter, r *http.Request) (string, error) {
+	switch distro.Get() {
+	case distro.Synology:
+		user, err := synoAuthn()
+		if err != nil {
+			http.Error(w, err.Error(), http.StatusUnauthorized)
+			return "", err
+		}
+		if err := authorizeSynology(user); err != nil {
+			http.Error(w, err.Error(), http.StatusForbidden)
+			return "", err
+		}
+		return user, nil
+	case distro.QNAP:
+		user, resp, err := qnapAuthn(r)
+		if err != nil {
+			http.Error(w, err.Error(), http.StatusUnauthorized)
+			return "", err
+		}
+		if resp.IsAdmin == 0 {
+			http.Error(w, err.Error(), http.StatusForbidden)
+			return "", err
+		}
+		return user, nil
+	}
 	return "", nil
 }
 
-func synoTokenRedirect(w http.ResponseWriter, r *http.Request) bool {
-	if distro.Get() != distro.Synology {
-		return false
+// authorizeSynology checks whether the provided user has access to the web UI
+// by consulting the membership of the "administrators" group.
+func authorizeSynology(name string) error {
+	yes, err := groupmember.IsMemberOfGroup("administrators", name)
+	if err != nil {
+		return err
 	}
+	if !yes {
+		return fmt.Errorf("not a member of administrators group")
+	}
+	return nil
+}
+
+type qnapAuthResponse struct {
+	AuthPassed int    `xml:"authPassed"`
+	IsAdmin    int    `xml:"isAdmin"`
+	AuthSID    string `xml:"authSid"`
+	ErrorValue int    `xml:"errorValue"`
+}
+
+func qnapAuthn(r *http.Request) (string, *qnapAuthResponse, error) {
+	user, err := r.Cookie("NAS_USER")
+	if err != nil {
+		return "", nil, err
+	}
+	token, err := r.Cookie("qtoken")
+	if err != nil {
+		return "", nil, err
+	}
+	query := url.Values{
+		"qtoken": []string{token.Value},
+		"user":   []string{user.Value},
+	}
+	u := url.URL{
+		Scheme:   r.URL.Scheme,
+		Host:     r.URL.Host,
+		Path:     "/cgi-bin/authLogin.cgi",
+		RawQuery: query.Encode(),
+	}
+	resp, err := http.Get(u.String())
+	if err != nil {
+		return "", nil, err
+	}
+	defer resp.Body.Close()
+	out, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		return "", nil, err
+	}
+	authResp := &qnapAuthResponse{}
+	if err := xml.Unmarshal(out, authResp); err != nil {
+		return "", nil, err
+	}
+	if authResp.AuthPassed == 0 {
+		return "", nil, fmt.Errorf("not authenticated")
+	}
+	return user.Value, authResp, nil
+}
+
+func synoAuthn() (string, error) {
+	cmd := exec.Command("/usr/syno/synoman/webman/modules/authenticate.cgi")
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		return "", fmt.Errorf("auth: %v: %s", err, out)
+	}
+	return strings.TrimSpace(string(out)), nil
+}
+
+func authRedirect(w http.ResponseWriter, r *http.Request) bool {
+	if distro.Get() == distro.Synology {
+		return synoTokenRedirect(w, r)
+	}
+	return false
+}
+
+func synoTokenRedirect(w http.ResponseWriter, r *http.Request) bool {
 	if r.Header.Get("X-Syno-Token") != "" {
 		return false
 	}
@@ -128,75 +250,13 @@ req.send(null);
 </body></html>
 `
 
-const authenticationRedirectHTML = `
-<html>
-<head>
-	<title>Redirecting...</title>
-	<style>
-		html,
-		body {
-			height: 100%;
-		}
-
-		html {
-			background-color: rgb(249, 247, 246);
-			font-family: ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol", "Noto Color Emoji";
-			line-height: 1.5;
-			-webkit-text-size-adjust: 100%;
-			-webkit-font-smoothing: antialiased;
-			-moz-osx-font-smoothing: grayscale;
-		}
-
-		body {
-			display: flex;
-			flex-direction: column;
-			align-items: center;
-			justify-content: center;
-		}
-
-		.spinner {
-			margin-bottom: 2rem;
-			border: 4px rgba(112, 110, 109, 0.5) solid;
-			border-left-color: transparent;
-			border-radius: 9999px;
-			width: 4rem;
-			height: 4rem;
-			-webkit-animation: spin 700ms linear infinite;
-      animation: spin 800ms linear infinite;
-		}
-
-		.label {
-			color: rgb(112, 110, 109);
-			padding-left: 0.4rem;
-		}
-
-		@-webkit-keyframes spin {
-			to {
-				transform: rotate(360deg);
-			}
-		}
-
-		@keyframes spin {
-			to {
-				transform: rotate(360deg);
-			}
-		}
-	</style>
-</head>
-<body>
-	<div class="spinner"></div>
-	<div class="label">Redirecting...</div>
-</body>
-`
-
 func webHandler(w http.ResponseWriter, r *http.Request) {
-	if synoTokenRedirect(w, r) {
+	if authRedirect(w, r) {
 		return
 	}
 
-	user, err := auth()
+	user, err := authorize(w, r)
 	if err != nil {
-		http.Error(w, err.Error(), http.StatusForbidden)
 		return
 	}
 
@@ -208,9 +268,10 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 	if r.Method == "POST" {
 		type mi map[string]interface{}
 		w.Header().Set("Content-Type", "application/json")
-		url, err := tailscaleUp(r.Context())
+		url, err := tailscaleUpForceReauth(r.Context())
 		if err != nil {
-			json.NewEncoder(w).Encode(mi{"error": err})
+			w.WriteHeader(http.StatusInternalServerError)
+			json.NewEncoder(w).Encode(mi{"error": err.Error()})
 			return
 		}
 		json.NewEncoder(w).Encode(mi{"url": url})
@@ -219,7 +280,7 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 
 	st, err := tailscale.Status(r.Context())
 	if err != nil {
-		http.Error(w, err.Error(), 500)
+		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 
@@ -237,16 +298,16 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 
 	buf := new(bytes.Buffer)
 	if err := tmpl.Execute(buf, data); err != nil {
-		http.Error(w, err.Error(), 500)
+		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 	w.Write(buf.Bytes())
 }
 
 // TODO(crawshaw): some of this is very similar to the code in 'tailscale up', can we share anything?
-func tailscaleUp(ctx context.Context) (authURL string, retErr error) {
+func tailscaleUpForceReauth(ctx context.Context) (authURL string, retErr error) {
 	prefs := ipn.NewPrefs()
-	prefs.ControlURL = "https://login.tailscale.com"
+	prefs.ControlURL = ipn.DefaultControlURL
 	prefs.WantRunning = true
 	prefs.CorpDNS = true
 	prefs.AllowSingleHosts = true
@@ -256,36 +317,70 @@ func tailscaleUp(ctx context.Context) (authURL string, retErr error) {
 		prefs.NetfilterMode = preftype.NetfilterOff
 	}
 
-	c, bc, ctx, cancel := connect(ctx)
+	st, err := tailscale.Status(ctx)
+	if err != nil {
+		return "", fmt.Errorf("can't fetch status: %v", err)
+	}
+	origAuthURL := st.AuthURL
+
+	// printAuthURL reports whether we should print out the
+	// provided auth URL from an IPN notify.
+	printAuthURL := func(url string) bool {
+		return url != origAuthURL
+	}
+
+	c, bc, pumpCtx, cancel := connect(ctx)
 	defer cancel()
 
+	gotEngineUpdate := make(chan bool, 1) // gets value upon an engine update
+	go pump(pumpCtx, bc, c)
+
+	bc.SetNotifyCallback(func(n ipn.Notify) {
+		if n.Engine != nil {
+			select {
+			case gotEngineUpdate <- true:
+			default:
+			}
+		}
+		if n.ErrMessage != nil {
+			msg := *n.ErrMessage
+			if msg == ipn.ErrMsgPermissionDenied {
+				switch runtime.GOOS {
+				case "windows":
+					msg += " (Tailscale service in use by other user?)"
+				default:
+					msg += " (try 'sudo tailscale up [...]')"
+				}
+			}
+			retErr = fmt.Errorf("backend error: %v", msg)
+			cancel()
+		} else if url := n.BrowseToURL; url != nil && printAuthURL(*url) {
+			authURL = *url
+			cancel()
+		}
+	})
+	// Wait for backend client to be connected so we know
+	// we're subscribed to updates. Otherwise we can miss
+	// an update upon its transition to running. Do so by causing some traffic
+	// back to the bus that we then wait on.
+	bc.RequestEngineStatus()
+	select {
+	case <-gotEngineUpdate:
+	case <-pumpCtx.Done():
+		return authURL, pumpCtx.Err()
+	}
+
 	bc.SetPrefs(prefs)
 
-	opts := ipn.Options{
+	bc.Start(ipn.Options{
 		StateKey: ipn.GlobalDaemonStateKey,
-		Notify: func(n ipn.Notify) {
-			if n.ErrMessage != nil {
-				msg := *n.ErrMessage
-				if msg == ipn.ErrMsgPermissionDenied {
-					switch runtime.GOOS {
-					case "windows":
-						msg += " (Tailscale service in use by other user?)"
-					default:
-						msg += " (try 'sudo tailscale up [...]')"
-					}
-				}
-				retErr = fmt.Errorf("backend error: %v", msg)
-				cancel()
-			} else if url := n.BrowseToURL; url != nil {
-				authURL = *url
-				cancel()
-			}
-		},
-	}
-	bc.Start(opts)
+	})
 	bc.StartLoginInteractive()
-	pump(ctx, bc, c)
 
+	<-pumpCtx.Done() // wait for authURL or complete failure
+	if authURL == "" && retErr == nil {
+		retErr = pumpCtx.Err()
+	}
 	if authURL == "" && retErr == nil {
 		return "", fmt.Errorf("login failed with no backend error message")
 	}

cmd/tailscale/cli/web.html

@@ -11,140 +11,133 @@
 </head>
 
 <body class="py-14">
-	<main class="container max-w-lg mx-auto py-6 px-8 bg-white rounded-md shadow-2xl" style="width: 95%">
-		<header class="flex justify-between items-center min-width-0 py-2 mb-8">
-			<svg width="26" height="26" viewBox="0 0 23 23" title="Tailscale" fill="none" xmlns="http://www.w3.org/2000/svg"
-				class="flex-shrink-0 mr-4">
-				<circle opacity="0.2" cx="3.4" cy="3.25" r="2.7" fill="currentColor"></circle>
-				<circle cx="3.4" cy="11.3" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="3.4" cy="19.5" r="2.7" fill="currentColor"></circle>
-				<circle cx="11.5" cy="11.3" r="2.7" fill="currentColor"></circle>
-				<circle cx="11.5" cy="19.5" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="11.5" cy="3.25" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="19.5" cy="3.25" r="2.7" fill="currentColor"></circle>
-				<circle cx="19.5" cy="11.3" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="19.5" cy="19.5" r="2.7" fill="currentColor"></circle>
-			</svg>
-			<div class="flex items-center justify-end space-x-2 w-2/3">
-				{{ with .Profile.LoginName }}
-				<div class="text-right truncate leading-4">
-					<h4 class="truncate">{{.}}</h4>
-					<a href="#" class="text-xs text-gray-500 hover:text-gray-700 js-loginButton">Switch account</a>
-				</div>
+<main class="container max-w-lg mx-auto py-6 px-8 bg-white rounded-md shadow-2xl" style="width: 95%">
+	<header class="flex justify-between items-center min-width-0 py-2 mb-8">
+		<svg width="26" height="26" viewBox="0 0 23 23" title="Tailscale" fill="none" xmlns="http://www.w3.org/2000/svg"
+			class="flex-shrink-0 mr-4">
+			<circle opacity="0.2" cx="3.4" cy="3.25" r="2.7" fill="currentColor"></circle>
+			<circle cx="3.4" cy="11.3" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="3.4" cy="19.5" r="2.7" fill="currentColor"></circle>
+			<circle cx="11.5" cy="11.3" r="2.7" fill="currentColor"></circle>
+			<circle cx="11.5" cy="19.5" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="11.5" cy="3.25" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="19.5" cy="3.25" r="2.7" fill="currentColor"></circle>
+			<circle cx="19.5" cy="11.3" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="19.5" cy="19.5" r="2.7" fill="currentColor"></circle>
+		</svg>
+		<div class="flex items-center justify-end space-x-2 w-2/3">
+			{{ with .Profile.LoginName }}
+			<div class="text-right truncate leading-4">
+				<h4 class="truncate">{{.}}</h4>
+				<a href="#" class="text-xs text-gray-500 hover:text-gray-700 js-loginButton">Switch account</a>
+			</div>
+			{{ end }}
+			<div class="relative flex-shrink-0 w-8 h-8 rounded-full overflow-hidden">
+				{{ with .Profile.ProfilePicURL }}
+				<div class="w-8 h-8 flex pointer-events-none rounded-full bg-gray-200"
+					style="background-image: url('{{.}}'); background-size: cover;"></div>
+				{{ else }}
+				<div class="w-8 h-8 flex pointer-events-none rounded-full border border-gray-400 border-dashed"></div>
 				{{ end }}
-				<div class="relative flex-shrink-0 w-8 h-8 rounded-full overflow-hidden">
-					{{ with .Profile.ProfilePicURL }}
-					<div class="w-8 h-8 flex pointer-events-none rounded-full bg-gray-200"
-						style="background-image: url('{{.}}'); background-size: cover;"></div>
-					{{ else }}
-					<div class="w-8 h-8 flex pointer-events-none rounded-full border border-gray-400 border-dashed"></div>
-					{{ end }}
-				</div>
 			</div>
-		</header>
-		{{ if .IP }}
-		<div
-			class="border border-gray-200 bg-gray-0 rounded-lg p-2 pl-3 pr-3 mb-8 width-full flex items-center justify-between">
-			<div class="flex items-center min-width-0">
-				<svg class="flex-shrink-0 text-gray-600 mr-3 ml-1" xmlns="http://www.w3.org/2000/svg" width="20" height="20"
-					viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"
-					stroke-linejoin="round">
-					<rect x="2" y="2" width="20" height="8" rx="2" ry="2"></rect>
-					<rect x="2" y="14" width="20" height="8" rx="2" ry="2"></rect>
-					<line x1="6" y1="6" x2="6.01" y2="6"></line>
-					<line x1="6" y1="18" x2="6.01" y2="18"></line>
-				</svg>
-				<h4 class="font-semibold truncate mr-2">{{.DeviceName}}</h4>
-			</div>
-			<h5>{{.IP}}</h5>
 		</div>
-		{{ end }}
-		{{ if or (eq .Status "NeedsLogin") (eq .Status "NoState") }}
-		{{ if .IP }}
-		<div class="mb-6">
-			<p class="text-gray-700">Your device's key has expired. Reauthenticate this device by logging in again, or <a
-					href="https://tailscale.com/kb/1028/key-expiry" class="link" target="_blank">learn more</a>.</p>
+	</header>
+	{{ if .IP }}
+	<div
+		class="border border-gray-200 bg-gray-0 rounded-lg p-2 pl-3 pr-3 mb-8 width-full flex items-center justify-between">
+		<div class="flex items-center min-width-0">
+			<svg class="flex-shrink-0 text-gray-600 mr-3 ml-1" xmlns="http://www.w3.org/2000/svg" width="20" height="20"
+				viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"
+				stroke-linejoin="round">
+				<rect x="2" y="2" width="20" height="8" rx="2" ry="2"></rect>
+				<rect x="2" y="14" width="20" height="8" rx="2" ry="2"></rect>
+				<line x1="6" y1="6" x2="6.01" y2="6"></line>
+				<line x1="6" y1="18" x2="6.01" y2="18"></line>
+			</svg>
+			<h4 class="font-semibold truncate mr-2">{{.DeviceName}}</h4>
 		</div>
-		<a href="#" class="mb-4 js-loginButton" target="_blank">
-			<button class="button button-blue w-full">Reauthenticate</button>
-		</a>
-		{{ else }}
-		<div class="mb-6">
-			<h3 class="text-3xl font-semibold mb-3">Log in</h3>
-			<p class="text-gray-700">Get started by logging in to your Tailscale network. Or,&nbsp;learn&nbsp;more at <a
-					href="https://tailscale.com/" class="link" target="_blank">tailscale.com</a>.</p>
-		</div>
-		<a href="#" class="mb-4 js-loginButton" target="_blank">
-			<button class="button button-blue w-full">Log In</button>
-		</a>
-		{{ end }}
-		{{ else if eq .Status "NeedsMachineAuth" }}
-		<div class="mb-4">
-			This device is authorized, but needs approval from a network admin before it can connect to the network.
-		</div>
-		{{ else }}
-		<div class="mb-4">
-			<p>You are connected! Access this device over Tailscale using the device name or IP address above.</p>
-		</div>
-		<a href="#" class="mb-4 link font-medium js-loginButton" target="_blank">Reauthenticate</a>
-		{{ end }}
-	</main>
-	<script>
-		(function () {
-			let loginButtons = document.querySelectorAll(".js-loginButton");
-			let fetchingUrl = false;
+		<h5>{{.IP}}</h5>
+	</div>
+	{{ end }}
+	{{ if or (eq .Status "NeedsLogin") (eq .Status "NoState") }}
+	{{ if .IP }}
+	<div class="mb-6">
+		<p class="text-gray-700">Your device's key has expired. Reauthenticate this device by logging in again, or <a
+				href="https://tailscale.com/kb/1028/key-expiry" class="link" target="_blank">learn more</a>.</p>
+	</div>
+	<a href="#" class="mb-4 js-loginButton" target="_blank">
+		<button class="button button-blue w-full">Reauthenticate</button>
+	</a>
+	{{ else }}
+	<div class="mb-6">
+		<h3 class="text-3xl font-semibold mb-3">Log in</h3>
+		<p class="text-gray-700">Get started by logging in to your Tailscale network. Or,&nbsp;learn&nbsp;more at <a
+				href="https://tailscale.com/" class="link" target="_blank">tailscale.com</a>.</p>
+	</div>
+	<a href="#" class="mb-4 js-loginButton" target="_blank">
+		<button class="button button-blue w-full">Log In</button>
+	</a>
+	{{ end }}
+	{{ else if eq .Status "NeedsMachineAuth" }}
+	<div class="mb-4">
+		This device is authorized, but needs approval from a network admin before it can connect to the network.
+	</div>
+	{{ else }}
+	<div class="mb-4">
+		<p>You are connected! Access this device over Tailscale using the device name or IP address above.</p>
+	</div>
+	<a href="#" class="mb-4 link font-medium js-loginButton" target="_blank">Reauthenticate</a>
+	{{ end }}
+</main>
+<script>(function () {
+let loginButtons = document.querySelectorAll(".js-loginButton");
+let fetchingUrl = false;
 
-			function handleClick(e) {
-				e.preventDefault();
+function handleClick(e) {
+	e.preventDefault();
 
-				if (fetchingUrl) {
-					return;
-				}
+	if (fetchingUrl) {
+		return;
+	}
 
-				fetchingUrl = true;
-				const urlParams = new URLSearchParams(window.location.search);
-				const token = urlParams.get("SynoToken");
-				const nextParams = new URLSearchParams({ up: true });
-				if (token) {
-					nextParams.set("SynoToken", token)
-				}
-				const nextUrl = new URL(window.location);
-				nextUrl.search = nextParams.toString()
-				const url = nextUrl.toString();
+	fetchingUrl = true;
+	const urlParams = new URLSearchParams(window.location.search);
+	const token = urlParams.get("SynoToken");
+	const nextParams = new URLSearchParams({ up: true });
+	if (token) {
+		nextParams.set("SynoToken", token)
+	}
+	const nextUrl = new URL(window.location);
+	nextUrl.search = nextParams.toString()
+	const url = nextUrl.toString();
 
-				const tab = window.open("/redirect", "_blank");
+	fetch(url, {
+		method: "POST",
+		headers: {
+			"Accept": "application/json",
+			"Content-Type": "application/json",
+		}
+	}).then(res => res.json()).then(res => {
+		fetchingUrl = false;
+		const err = res["error"];
+		if (err) {
+			throw new Error(err);
+		}
+		const url = res["url"];
+		if (url) {
+			document.location.href = url;
+		} else {
+			location.reload();
+		}
+	}).catch(err => {
+		alert("Failed to log in: " + err.message);
+	});
+}
 
-				fetch(url, {
-					method: "POST",
-					headers: {
-						"Accept": "application/json",
-						"Content-Type": "application/json",
-					}
-				}).then(res => res.json()).then(res => {
-					fetchingUrl = false;
-					const err = res["error"];
-					if (err) {
-						throw new Error(err);
-					}
-					const url = res["url"];
-					if (url) {
-						authUrl = url;
-						tab.location = url;
-						tab.focus();
-					} else {
-						location.reload();
-					}
-				}).catch(err => {
-					tab.close();
-					alert("Failed to log in: " + err.message);
-				});
-			}
-
-			Array.from(loginButtons).forEach(el => {
-				el.addEventListener("click", handleClick);
-			})
-		})();
-	</script>
+Array.from(loginButtons).forEach(el => {
+	el.addEventListener("click", handleClick);
+})
+})();</script>
 </body>
 
 </html>

cmd/tailscale/cli/web_test.go

@@ -0,0 +1,44 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import "testing"
+
+func TestUrlOfListenAddr(t *testing.T) {
+	tests := []struct {
+		name     string
+		in, want string
+	}{
+		{
+			name: "TestLocalhost",
+			in:   "localhost:8088",
+			want: "http://localhost:8088",
+		},
+		{
+			name: "TestNoHost",
+			in:   ":8088",
+			want: "http://127.0.0.1:8088",
+		},
+		{
+			name: "TestExplicitHost",
+			in:   "127.0.0.2:8088",
+			want: "http://127.0.0.2:8088",
+		},
+		{
+			name: "TestIPv6",
+			in:   "[::1]:8088",
+			want: "http://[::1]:8088",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			url := urlOfListenAddr(tt.in)
+			if url != tt.want {
+				t.Errorf("expected url: %q, got: %q", tt.want, url)
+			}
+		})
+	}
+}

cmd/tailscale/depaware.txt

@@ -1,9 +1,19 @@
 tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/depaware)
 
-   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate
+   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate+
+   W    github.com/alexbrainman/sspi/internal/common                 from github.com/alexbrainman/sspi/negotiate
    W 💣 github.com/alexbrainman/sspi/negotiate                       from tailscale.com/net/tshttpproxy
+        github.com/dsnet/golib/jsonfmt                               from tailscale.com/cmd/tailscale/cli
+        github.com/kballard/go-shellquote                            from tailscale.com/cmd/tailscale/cli
+     💣 github.com/mitchellh/go-ps                                   from tailscale.com/cmd/tailscale/cli
         github.com/peterbourgon/ff/v2                                from github.com/peterbourgon/ff/v2/ffcli
         github.com/peterbourgon/ff/v2/ffcli                          from tailscale.com/cmd/tailscale/cli
+        github.com/tailscale/goupnp                                  from github.com/tailscale/goupnp/dcps/internetgateway2+
+        github.com/tailscale/goupnp/dcps/internetgateway2            from tailscale.com/net/portmapper
+        github.com/tailscale/goupnp/httpu                            from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/scpd                             from github.com/tailscale/goupnp
+        github.com/tailscale/goupnp/soap                             from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/ssdp                             from github.com/tailscale/goupnp
         github.com/tcnksm/go-httpstat                                from tailscale.com/net/netcheck
         github.com/toqueteos/webbrowser                              from tailscale.com/cmd/tailscale/cli
      💣 go4.org/intern                                               from inet.af/netaddr
@@ -13,13 +23,15 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         inet.af/netaddr                                              from tailscale.com/cmd/tailscale/cli+
         rsc.io/goversion/version                                     from tailscale.com/version
         tailscale.com/atomicfile                                     from tailscale.com/ipn
-        tailscale.com/client/tailscale                               from tailscale.com/cmd/tailscale/cli
+        tailscale.com/client/tailscale                               from tailscale.com/cmd/tailscale/cli+
+        tailscale.com/client/tailscale/apitype                       from tailscale.com/client/tailscale+
         tailscale.com/cmd/tailscale/cli                              from tailscale.com/cmd/tailscale
+        tailscale.com/control/controlknobs                           from tailscale.com/net/portmapper
         tailscale.com/derp                                           from tailscale.com/derp/derphttp
         tailscale.com/derp/derphttp                                  from tailscale.com/net/netcheck
-        tailscale.com/derp/derpmap                                   from tailscale.com/cmd/tailscale/cli
         tailscale.com/disco                                          from tailscale.com/derp
-        tailscale.com/ipn                                            from tailscale.com/cmd/tailscale/cli
+        tailscale.com/hostinfo                                       from tailscale.com/net/interfaces
+        tailscale.com/ipn                                            from tailscale.com/cmd/tailscale/cli+
         tailscale.com/ipn/ipnstate                                   from tailscale.com/cmd/tailscale/cli+
         tailscale.com/metrics                                        from tailscale.com/derp
         tailscale.com/net/dnscache                                   from tailscale.com/derp/derphttp
@@ -31,29 +43,33 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/net/portmapper                                 from tailscale.com/net/netcheck+
         tailscale.com/net/stun                                       from tailscale.com/net/netcheck
         tailscale.com/net/tlsdial                                    from tailscale.com/derp/derphttp
-        tailscale.com/net/tsaddr                                     from tailscale.com/net/interfaces
+        tailscale.com/net/tsaddr                                     from tailscale.com/net/interfaces+
      💣 tailscale.com/net/tshttpproxy                                from tailscale.com/derp/derphttp+
         tailscale.com/paths                                          from tailscale.com/cmd/tailscale/cli+
         tailscale.com/safesocket                                     from tailscale.com/cmd/tailscale/cli+
         tailscale.com/syncs                                          from tailscale.com/net/interfaces+
         tailscale.com/tailcfg                                        from tailscale.com/cmd/tailscale/cli+
    W    tailscale.com/tsconst                                        from tailscale.com/net/interfaces
+     💣 tailscale.com/tstime/mono                                    from tailscale.com/tstime/rate
+        tailscale.com/tstime/rate                                    from tailscale.com/wgengine/filter
+        tailscale.com/types/dnstype                                  from tailscale.com/tailcfg
         tailscale.com/types/empty                                    from tailscale.com/ipn
         tailscale.com/types/ipproto                                  from tailscale.com/net/flowtrack+
         tailscale.com/types/key                                      from tailscale.com/derp+
         tailscale.com/types/logger                                   from tailscale.com/cmd/tailscale/cli+
         tailscale.com/types/netmap                                   from tailscale.com/ipn
         tailscale.com/types/opt                                      from tailscale.com/net/netcheck+
+        tailscale.com/types/pad32                                    from tailscale.com/derp
         tailscale.com/types/persist                                  from tailscale.com/ipn
         tailscale.com/types/preftype                                 from tailscale.com/cmd/tailscale/cli+
-        tailscale.com/types/strbuilder                               from tailscale.com/net/packet
         tailscale.com/types/structs                                  from tailscale.com/ipn+
         tailscale.com/types/wgkey                                    from tailscale.com/types/netmap+
         tailscale.com/util/dnsname                                   from tailscale.com/cmd/tailscale/cli+
    W    tailscale.com/util/endian                                    from tailscale.com/net/netns
-   L    tailscale.com/util/lineread                                  from tailscale.com/net/interfaces
+        tailscale.com/util/groupmember                               from tailscale.com/cmd/tailscale/cli
+        tailscale.com/util/lineread                                  from tailscale.com/net/interfaces+
         tailscale.com/version                                        from tailscale.com/cmd/tailscale/cli+
-        tailscale.com/version/distro                                 from tailscale.com/cmd/tailscale/cli
+        tailscale.com/version/distro                                 from tailscale.com/cmd/tailscale/cli+
         tailscale.com/wgengine/filter                                from tailscale.com/types/netmap
         golang.org/x/crypto/blake2b                                  from golang.org/x/crypto/nacl/box
         golang.org/x/crypto/chacha20                                 from golang.org/x/crypto/chacha20poly1305
@@ -73,7 +89,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         golang.org/x/net/idna                                        from golang.org/x/net/http/httpguts+
         golang.org/x/net/proxy                                       from tailscale.com/net/netns
    D    golang.org/x/net/route                                       from net+
-        golang.org/x/sync/errgroup                                   from tailscale.com/derp
+        golang.org/x/sync/errgroup                                   from tailscale.com/derp+
         golang.org/x/sync/singleflight                               from tailscale.com/net/dnscache
         golang.org/x/sys/cpu                                         from golang.org/x/crypto/blake2b+
   LD    golang.org/x/sys/unix                                        from tailscale.com/net/netns+
@@ -83,7 +99,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         golang.org/x/text/transform                                  from golang.org/x/text/secure/bidirule+
         golang.org/x/text/unicode/bidi                               from golang.org/x/net/idna+
         golang.org/x/text/unicode/norm                               from golang.org/x/net/idna
-        golang.org/x/time/rate                                       from tailscale.com/types/logger+
+        golang.org/x/time/rate                                       from tailscale.com/cmd/tailscale/cli+
         bufio                                                        from compress/flate+
         bytes                                                        from bufio+
         compress/flate                                               from compress/gzip+
@@ -116,13 +132,14 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         debug/macho                                                  from rsc.io/goversion/version
         debug/pe                                                     from rsc.io/goversion/version
         embed                                                        from tailscale.com/cmd/tailscale/cli
-        encoding                                                     from encoding/json
+        encoding                                                     from encoding/json+
         encoding/asn1                                                from crypto/x509+
         encoding/base64                                              from encoding/json+
         encoding/binary                                              from compress/gzip+
         encoding/hex                                                 from crypto/x509+
         encoding/json                                                from expvar+
         encoding/pem                                                 from crypto/tls+
+        encoding/xml                                                 from tailscale.com/cmd/tailscale/cli+
         errors                                                       from bufio+
         expvar                                                       from tailscale.com/derp+
         flag                                                         from github.com/peterbourgon/ff/v2+
@@ -154,6 +171,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         os                                                           from crypto/rand+
         os/exec                                                      from github.com/toqueteos/webbrowser+
         os/signal                                                    from tailscale.com/cmd/tailscale/cli
+        os/user                                                      from tailscale.com/util/groupmember
         path                                                         from debug/dwarf+
         path/filepath                                                from crypto/x509+
         reflect                                                      from crypto/x509+

cmd/tailscaled/debug.go

@@ -11,19 +11,26 @@ import (
 	"errors"
 	"flag"
 	"fmt"
+	"io"
+	"io/ioutil"
 	"log"
+	"net"
 	"net/http"
 	"net/http/httptrace"
 	"net/url"
 	"os"
+	"strings"
 	"time"
 
+	"inet.af/netaddr"
 	"tailscale.com/derp/derphttp"
-	"tailscale.com/derp/derpmap"
+	"tailscale.com/ipn"
 	"tailscale.com/net/interfaces"
+	"tailscale.com/net/portmapper"
 	"tailscale.com/net/tshttpproxy"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
 	"tailscale.com/wgengine/monitor"
 )
 
@@ -31,6 +38,7 @@ var debugArgs struct {
 	monitor   bool
 	getURL    string
 	derpCheck string
+	portmap   bool
 }
 
 var debugModeFunc = debugMode // so it can be addressable
@@ -38,6 +46,7 @@ var debugModeFunc = debugMode // so it can be addressable
 func debugMode(args []string) error {
 	fs := flag.NewFlagSet("debug", flag.ExitOnError)
 	fs.BoolVar(&debugArgs.monitor, "monitor", false, "If true, run link monitor forever. Precludes all other options.")
+	fs.BoolVar(&debugArgs.portmap, "portmap", false, "If true, run portmap debugging. Precludes all other options.")
 	fs.StringVar(&debugArgs.getURL, "get-url", "", "If non-empty, fetch provided URL.")
 	fs.StringVar(&debugArgs.derpCheck, "derp", "", "if non-empty, test a DERP ping via named region code")
 	if err := fs.Parse(args); err != nil {
@@ -53,6 +62,9 @@ func debugMode(args []string) error {
 	if debugArgs.monitor {
 		return runMonitor(ctx)
 	}
+	if debugArgs.portmap {
+		return debugPortmap(ctx)
+	}
 	if debugArgs.getURL != "" {
 		return getURL(ctx, debugArgs.getURL)
 	}
@@ -131,7 +143,26 @@ func getURL(ctx context.Context, urlStr string) error {
 }
 
 func checkDerp(ctx context.Context, derpRegion string) error {
-	dmap := derpmap.Prod()
+	req, err := http.NewRequestWithContext(ctx, "GET", ipn.DefaultControlURL+"/derpmap/default", nil)
+	if err != nil {
+		return fmt.Errorf("create derp map request: %w", err)
+	}
+	res, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("fetch derp map failed: %w", err)
+	}
+	defer res.Body.Close()
+	b, err := ioutil.ReadAll(io.LimitReader(res.Body, 1<<20))
+	if err != nil {
+		return fmt.Errorf("fetch derp map failed: %w", err)
+	}
+	if res.StatusCode != 200 {
+		return fmt.Errorf("fetch derp map: %v: %s", res.Status, b)
+	}
+	var dmap tailcfg.DERPMap
+	if err = json.Unmarshal(b, &dmap); err != nil {
+		return fmt.Errorf("fetch DERP map: %w", err)
+	}
 	getRegion := func() *tailcfg.DERPRegion {
 		for _, r := range dmap.Regions {
 			if r.RegionCode == derpRegion {
@@ -170,3 +201,97 @@ func checkDerp(ctx context.Context, derpRegion string) error {
 	log.Printf("ok")
 	return err
 }
+
+func debugPortmap(ctx context.Context) error {
+	ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+	defer cancel()
+
+	portmapper.VerboseLogs = true
+	switch os.Getenv("TS_DEBUG_PORTMAP_TYPE") {
+	case "":
+	case "pmp":
+		portmapper.DisablePCP = true
+		portmapper.DisableUPnP = true
+	case "pcp":
+		portmapper.DisablePMP = true
+		portmapper.DisableUPnP = true
+	case "upnp":
+		portmapper.DisablePCP = true
+		portmapper.DisablePMP = true
+	default:
+		log.Fatalf("TS_DEBUG_PORTMAP_TYPE must be one of pmp,pcp,upnp")
+	}
+
+	done := make(chan bool, 1)
+
+	var c *portmapper.Client
+	logf := log.Printf
+	c = portmapper.NewClient(logger.WithPrefix(logf, "portmapper: "), func() {
+		logf("portmapping changed.")
+		logf("have mapping: %v", c.HaveMapping())
+
+		if ext, ok := c.GetCachedMappingOrStartCreatingOne(); ok {
+			logf("cb: mapping: %v", ext)
+			select {
+			case done <- true:
+			default:
+			}
+			return
+		}
+		logf("cb: no mapping")
+	})
+	linkMon, err := monitor.New(logger.WithPrefix(logf, "monitor: "))
+	if err != nil {
+		return err
+	}
+
+	gatewayAndSelfIP := func() (gw, self netaddr.IP, ok bool) {
+		if v := os.Getenv("TS_DEBUG_GW_SELF"); strings.Contains(v, "/") {
+			i := strings.Index(v, "/")
+			gw = netaddr.MustParseIP(v[:i])
+			self = netaddr.MustParseIP(v[i+1:])
+			return gw, self, true
+		}
+		return linkMon.GatewayAndSelfIP()
+	}
+
+	c.SetGatewayLookupFunc(gatewayAndSelfIP)
+
+	gw, selfIP, ok := gatewayAndSelfIP()
+	if !ok {
+		logf("no gateway or self IP; %v", linkMon.InterfaceState())
+		return nil
+	}
+	logf("gw=%v; self=%v", gw, selfIP)
+
+	uc, err := net.ListenPacket("udp", "0.0.0.0:0")
+	if err != nil {
+		return err
+	}
+	defer uc.Close()
+	c.SetLocalPort(uint16(uc.LocalAddr().(*net.UDPAddr).Port))
+
+	res, err := c.Probe(ctx)
+	if err != nil {
+		return fmt.Errorf("Probe: %v", err)
+	}
+	logf("Probe: %+v", res)
+
+	if !res.PCP && !res.PMP && !res.UPnP {
+		logf("no portmapping services available")
+		return nil
+	}
+
+	if ext, ok := c.GetCachedMappingOrStartCreatingOne(); ok {
+		logf("mapping: %v", ext)
+	} else {
+		logf("no mapping")
+	}
+
+	select {
+	case <-done:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}

cmd/tailscaled/depaware.txt

@@ -1,43 +1,60 @@
 tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/depaware)
 
-   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate
+   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate+
+   W    github.com/alexbrainman/sspi/internal/common                 from github.com/alexbrainman/sspi/negotiate
    W 💣 github.com/alexbrainman/sspi/negotiate                       from tailscale.com/net/tshttpproxy
    L    github.com/coreos/go-iptables/iptables                       from tailscale.com/wgengine/router
-   W 💣 github.com/github/certstore                                  from tailscale.com/control/controlclient
         github.com/go-multierror/multierror                          from tailscale.com/wgengine/router+
    W 💣 github.com/go-ole/go-ole                                     from github.com/go-ole/go-ole/oleutil+
    W 💣 github.com/go-ole/go-ole/oleutil                             from tailscale.com/wgengine/winnet
    L 💣 github.com/godbus/dbus/v5                                    from tailscale.com/net/dns
+        github.com/golang/snappy                                     from github.com/klauspost/compress/zstd
         github.com/google/btree                                      from inet.af/netstack/tcpip/header+
+   L    github.com/insomniacslk/dhcp/dhcpv4                          from tailscale.com/net/tstun
+   L    github.com/insomniacslk/dhcp/iana                            from github.com/insomniacslk/dhcp/dhcpv4
+   L    github.com/insomniacslk/dhcp/interfaces                      from github.com/insomniacslk/dhcp/dhcpv4
+   L    github.com/insomniacslk/dhcp/rfc1035label                    from github.com/insomniacslk/dhcp/dhcpv4
    L    github.com/josharian/native                                  from github.com/mdlayher/netlink+
    L 💣 github.com/jsimonetti/rtnetlink                              from tailscale.com/wgengine/monitor
    L    github.com/jsimonetti/rtnetlink/internal/unix                from github.com/jsimonetti/rtnetlink
         github.com/klauspost/compress/fse                            from github.com/klauspost/compress/huff0
         github.com/klauspost/compress/huff0                          from github.com/klauspost/compress/zstd
-        github.com/klauspost/compress/snappy                         from github.com/klauspost/compress/zstd
         github.com/klauspost/compress/zstd                           from tailscale.com/smallzstd
         github.com/klauspost/compress/zstd/internal/xxhash           from github.com/klauspost/compress/zstd
-   L 💣 github.com/mdlayher/netlink                                  from github.com/jsimonetti/rtnetlink+
-   L 💣 github.com/mdlayher/netlink/nlenc                            from github.com/jsimonetti/rtnetlink+
+   L 💣 github.com/mdlayher/netlink                                  from tailscale.com/wgengine/monitor+
+   L 💣 github.com/mdlayher/netlink/nlenc                            from github.com/mdlayher/netlink+
    L    github.com/mdlayher/sdnotify                                 from tailscale.com/util/systemd
-   W    github.com/pkg/errors                                        from github.com/github/certstore
-     💣 github.com/tailscale/wireguard-go/conn                       from github.com/tailscale/wireguard-go/device+
-   W 💣 github.com/tailscale/wireguard-go/conn/winrio                from github.com/tailscale/wireguard-go/conn
-     💣 github.com/tailscale/wireguard-go/device                     from tailscale.com/wgengine+
-     💣 github.com/tailscale/wireguard-go/ipc                        from github.com/tailscale/wireguard-go/device
-   W 💣 github.com/tailscale/wireguard-go/ipc/winpipe                from github.com/tailscale/wireguard-go/ipc
-        github.com/tailscale/wireguard-go/ratelimiter                from github.com/tailscale/wireguard-go/device
-        github.com/tailscale/wireguard-go/replay                     from github.com/tailscale/wireguard-go/device
-        github.com/tailscale/wireguard-go/rwcancel                   from github.com/tailscale/wireguard-go/device+
-        github.com/tailscale/wireguard-go/tai64n                     from github.com/tailscale/wireguard-go/device+
-     💣 github.com/tailscale/wireguard-go/tun                        from github.com/tailscale/wireguard-go/device+
-   W 💣 github.com/tailscale/wireguard-go/tun/wintun                 from github.com/tailscale/wireguard-go/tun+
+   L 💣 github.com/mdlayher/socket                                   from github.com/mdlayher/netlink
+   W    github.com/pkg/errors                                        from github.com/tailscale/certstore
+   W 💣 github.com/tailscale/certstore                               from tailscale.com/control/controlclient
+        github.com/tailscale/goupnp                                  from github.com/tailscale/goupnp/dcps/internetgateway2+
+        github.com/tailscale/goupnp/dcps/internetgateway2            from tailscale.com/net/portmapper
+        github.com/tailscale/goupnp/httpu                            from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/scpd                             from github.com/tailscale/goupnp
+        github.com/tailscale/goupnp/soap                             from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/ssdp                             from github.com/tailscale/goupnp
         github.com/tcnksm/go-httpstat                                from tailscale.com/net/netcheck
+   L    github.com/u-root/uio/rand                                   from github.com/insomniacslk/dhcp/dhcpv4
+   L    github.com/u-root/uio/ubinary                                from github.com/u-root/uio/uio
+   L    github.com/u-root/uio/uio                                    from github.com/insomniacslk/dhcp/dhcpv4+
      💣 go4.org/intern                                               from inet.af/netaddr
-     💣 go4.org/mem                                                  from tailscale.com/control/controlclient+
+     💣 go4.org/mem                                                  from tailscale.com/derp+
         go4.org/unsafe/assume-no-moving-gc                           from go4.org/intern
+     💣 golang.zx2c4.com/wireguard/conn                              from golang.zx2c4.com/wireguard/device+
+   W 💣 golang.zx2c4.com/wireguard/conn/winrio                       from golang.zx2c4.com/wireguard/conn
+     💣 golang.zx2c4.com/wireguard/device                            from tailscale.com/net/tstun+
+     💣 golang.zx2c4.com/wireguard/ipc                               from golang.zx2c4.com/wireguard/device
+   W 💣 golang.zx2c4.com/wireguard/ipc/winpipe                       from golang.zx2c4.com/wireguard/ipc
+        golang.zx2c4.com/wireguard/ratelimiter                       from golang.zx2c4.com/wireguard/device
+        golang.zx2c4.com/wireguard/replay                            from golang.zx2c4.com/wireguard/device
+        golang.zx2c4.com/wireguard/rwcancel                          from golang.zx2c4.com/wireguard/device+
+        golang.zx2c4.com/wireguard/tai64n                            from golang.zx2c4.com/wireguard/device+
+     💣 golang.zx2c4.com/wireguard/tun                               from golang.zx2c4.com/wireguard/device+
+   W 💣 golang.zx2c4.com/wireguard/tun/wintun                        from golang.zx2c4.com/wireguard/tun+
    W 💣 golang.zx2c4.com/wireguard/windows/tunnel/winipcfg           from tailscale.com/net/interfaces+
         inet.af/netaddr                                              from tailscale.com/control/controlclient+
+        inet.af/netstack/atomicbitops                                from inet.af/netstack/tcpip+
+     💣 inet.af/netstack/buffer                                      from inet.af/netstack/tcpip/stack
      💣 inet.af/netstack/gohacks                                     from inet.af/netstack/state/wire+
         inet.af/netstack/linewriter                                  from inet.af/netstack/log
         inet.af/netstack/log                                         from inet.af/netstack/state+
@@ -46,7 +63,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
      💣 inet.af/netstack/state                                       from inet.af/netstack/tcpip+
         inet.af/netstack/state/wire                                  from inet.af/netstack/state
      💣 inet.af/netstack/sync                                        from inet.af/netstack/linewriter+
-     💣 inet.af/netstack/tcpip                                       from inet.af/netstack/tcpip/adapters/gonet+
+        inet.af/netstack/tcpip                                       from inet.af/netstack/tcpip/adapters/gonet+
         inet.af/netstack/tcpip/adapters/gonet                        from tailscale.com/wgengine/netstack
      💣 inet.af/netstack/tcpip/buffer                                from inet.af/netstack/tcpip/adapters/gonet+
         inet.af/netstack/tcpip/hash/jenkins                          from inet.af/netstack/tcpip/stack+
@@ -56,7 +73,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         inet.af/netstack/tcpip/network/hash                          from inet.af/netstack/tcpip/network/ipv4+
         inet.af/netstack/tcpip/network/internal/fragmentation        from inet.af/netstack/tcpip/network/ipv4+
         inet.af/netstack/tcpip/network/internal/ip                   from inet.af/netstack/tcpip/network/ipv4+
-        inet.af/netstack/tcpip/network/ipv4                          from tailscale.com/wgengine/netstack
+        inet.af/netstack/tcpip/network/ipv4                          from tailscale.com/wgengine/netstack+
         inet.af/netstack/tcpip/network/ipv6                          from tailscale.com/wgengine/netstack
         inet.af/netstack/tcpip/ports                                 from inet.af/netstack/tcpip/stack+
         inet.af/netstack/tcpip/seqnum                                from inet.af/netstack/tcpip/header+
@@ -69,15 +86,18 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         inet.af/netstack/tcpip/transport/udp                         from inet.af/netstack/tcpip/adapters/gonet+
         inet.af/netstack/waiter                                      from inet.af/netstack/tcpip+
         inet.af/peercred                                             from tailscale.com/ipn/ipnserver
+   W 💣 inet.af/wf                                                   from tailscale.com/wf
         rsc.io/goversion/version                                     from tailscale.com/version
         tailscale.com/atomicfile                                     from tailscale.com/ipn+
+        tailscale.com/client/tailscale                               from tailscale.com/derp
+        tailscale.com/client/tailscale/apitype                       from tailscale.com/ipn/ipnlocal+
         tailscale.com/control/controlclient                          from tailscale.com/ipn/ipnlocal+
+        tailscale.com/control/controlknobs                           from tailscale.com/control/controlclient+
         tailscale.com/derp                                           from tailscale.com/derp/derphttp+
         tailscale.com/derp/derphttp                                  from tailscale.com/net/netcheck+
-        tailscale.com/derp/derpmap                                   from tailscale.com/cmd/tailscaled+
         tailscale.com/disco                                          from tailscale.com/derp+
         tailscale.com/health                                         from tailscale.com/control/controlclient+
-        tailscale.com/internal/deepprint                             from tailscale.com/ipn/ipnlocal+
+        tailscale.com/hostinfo                                       from tailscale.com/control/controlclient+
         tailscale.com/ipn                                            from tailscale.com/ipn/ipnserver+
         tailscale.com/ipn/ipnlocal                                   from tailscale.com/ipn/ipnserver+
         tailscale.com/ipn/ipnserver                                  from tailscale.com/cmd/tailscaled
@@ -102,21 +122,24 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
      💣 tailscale.com/net/netstat                                    from tailscale.com/ipn/ipnserver
         tailscale.com/net/packet                                     from tailscale.com/wgengine+
         tailscale.com/net/portmapper                                 from tailscale.com/net/netcheck+
-        tailscale.com/net/socks5                                     from tailscale.com/cmd/tailscaled
+        tailscale.com/net/socks5                                     from tailscale.com/net/socks5/tssocks
+        tailscale.com/net/socks5/tssocks                             from tailscale.com/cmd/tailscaled
         tailscale.com/net/stun                                       from tailscale.com/net/netcheck+
         tailscale.com/net/tlsdial                                    from tailscale.com/control/controlclient+
         tailscale.com/net/tsaddr                                     from tailscale.com/ipn/ipnlocal+
      💣 tailscale.com/net/tshttpproxy                                from tailscale.com/control/controlclient+
-        tailscale.com/net/tstun                                      from tailscale.com/cmd/tailscaled+
+     💣 tailscale.com/net/tstun                                      from tailscale.com/cmd/tailscaled+
         tailscale.com/paths                                          from tailscale.com/cmd/tailscaled+
         tailscale.com/portlist                                       from tailscale.com/ipn/ipnlocal
-        tailscale.com/safesocket                                     from tailscale.com/ipn/ipnserver
+        tailscale.com/safesocket                                     from tailscale.com/ipn/ipnserver+
         tailscale.com/smallzstd                                      from tailscale.com/ipn/ipnserver+
         tailscale.com/syncs                                          from tailscale.com/net/interfaces+
         tailscale.com/tailcfg                                        from tailscale.com/control/controlclient+
-   W 💣 tailscale.com/tempfork/wireguard-windows/firewall            from tailscale.com/cmd/tailscaled
    W    tailscale.com/tsconst                                        from tailscale.com/net/interfaces
         tailscale.com/tstime                                         from tailscale.com/wgengine/magicsock
+     💣 tailscale.com/tstime/mono                                    from tailscale.com/net/tstun+
+        tailscale.com/tstime/rate                                    from tailscale.com/wgengine/filter
+        tailscale.com/types/dnstype                                  from tailscale.com/ipn/ipnlocal+
         tailscale.com/types/empty                                    from tailscale.com/control/controlclient+
         tailscale.com/types/flagtype                                 from tailscale.com/cmd/tailscaled
         tailscale.com/types/ipproto                                  from tailscale.com/net/flowtrack+
@@ -125,32 +148,38 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/types/netmap                                   from tailscale.com/control/controlclient+
         tailscale.com/types/nettype                                  from tailscale.com/wgengine/magicsock
         tailscale.com/types/opt                                      from tailscale.com/control/controlclient+
+        tailscale.com/types/pad32                                    from tailscale.com/net/tstun+
         tailscale.com/types/persist                                  from tailscale.com/control/controlclient+
         tailscale.com/types/preftype                                 from tailscale.com/ipn+
-        tailscale.com/types/strbuilder                               from tailscale.com/net/packet
         tailscale.com/types/structs                                  from tailscale.com/control/controlclient+
         tailscale.com/types/wgkey                                    from tailscale.com/control/controlclient+
+   L    tailscale.com/util/cmpver                                    from tailscale.com/net/dns
+     💣 tailscale.com/util/deephash                                  from tailscale.com/ipn/ipnlocal+
         tailscale.com/util/dnsname                                   from tailscale.com/ipn/ipnstate+
   LW    tailscale.com/util/endian                                    from tailscale.com/net/netns+
-   L    tailscale.com/util/lineread                                  from tailscale.com/control/controlclient+
+        tailscale.com/util/groupmember                               from tailscale.com/ipn/ipnserver
+        tailscale.com/util/lineread                                  from tailscale.com/control/controlclient+
+        tailscale.com/util/osshare                                   from tailscale.com/cmd/tailscaled+
         tailscale.com/util/pidowner                                  from tailscale.com/ipn/ipnserver
         tailscale.com/util/racebuild                                 from tailscale.com/logpolicy
         tailscale.com/util/systemd                                   from tailscale.com/control/controlclient+
+        tailscale.com/util/uniq                                      from tailscale.com/wgengine/magicsock
         tailscale.com/util/winutil                                   from tailscale.com/logpolicy+
         tailscale.com/version                                        from tailscale.com/cmd/tailscaled+
         tailscale.com/version/distro                                 from tailscale.com/control/controlclient+
+   W    tailscale.com/wf                                             from tailscale.com/cmd/tailscaled
         tailscale.com/wgengine                                       from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/filter                                from tailscale.com/control/controlclient+
-        tailscale.com/wgengine/magicsock                             from tailscale.com/cmd/tailscaled+
+        tailscale.com/wgengine/magicsock                             from tailscale.com/wgengine+
         tailscale.com/wgengine/monitor                               from tailscale.com/wgengine+
-        tailscale.com/wgengine/netstack                              from tailscale.com/cmd/tailscaled
+        tailscale.com/wgengine/netstack                              from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/router                                from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/wgcfg                                 from tailscale.com/ipn/ipnlocal+
         tailscale.com/wgengine/wgcfg/nmcfg                           from tailscale.com/ipn/ipnlocal
         tailscale.com/wgengine/wglog                                 from tailscale.com/wgengine
    W 💣 tailscale.com/wgengine/winnet                                from tailscale.com/wgengine/router
         golang.org/x/crypto/blake2b                                  from golang.org/x/crypto/nacl/box
-        golang.org/x/crypto/blake2s                                  from github.com/tailscale/wireguard-go/device+
+        golang.org/x/crypto/blake2s                                  from golang.zx2c4.com/wireguard/device+
         golang.org/x/crypto/chacha20                                 from golang.org/x/crypto/chacha20poly1305
         golang.org/x/crypto/chacha20poly1305                         from crypto/tls+
         golang.org/x/crypto/cryptobyte                               from crypto/ecdsa+
@@ -159,31 +188,32 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         golang.org/x/crypto/hkdf                                     from crypto/tls
         golang.org/x/crypto/nacl/box                                 from tailscale.com/control/controlclient+
         golang.org/x/crypto/nacl/secretbox                           from golang.org/x/crypto/nacl/box
-        golang.org/x/crypto/poly1305                                 from github.com/tailscale/wireguard-go/device+
+        golang.org/x/crypto/poly1305                                 from golang.org/x/crypto/chacha20poly1305+
         golang.org/x/crypto/salsa20/salsa                            from golang.org/x/crypto/nacl/box+
         golang.org/x/net/bpf                                         from github.com/mdlayher/netlink+
         golang.org/x/net/dns/dnsmessage                              from net+
-        golang.org/x/net/http/httpguts                               from net/http
+        golang.org/x/net/http/httpguts                               from net/http+
         golang.org/x/net/http/httpproxy                              from net/http
         golang.org/x/net/http2/hpack                                 from net/http
         golang.org/x/net/idna                                        from golang.org/x/net/http/httpguts+
-        golang.org/x/net/ipv4                                        from github.com/tailscale/wireguard-go/device
-        golang.org/x/net/ipv6                                        from github.com/tailscale/wireguard-go/device+
+        golang.org/x/net/ipv4                                        from golang.zx2c4.com/wireguard/device
+        golang.org/x/net/ipv6                                        from golang.zx2c4.com/wireguard/device+
         golang.org/x/net/proxy                                       from tailscale.com/net/netns
    D    golang.org/x/net/route                                       from net+
-        golang.org/x/sync/errgroup                                   from tailscale.com/derp
+        golang.org/x/sync/errgroup                                   from tailscale.com/derp+
         golang.org/x/sync/singleflight                               from tailscale.com/net/dnscache
         golang.org/x/sys/cpu                                         from golang.org/x/crypto/blake2b+
-  LD    golang.org/x/sys/unix                                        from github.com/jsimonetti/rtnetlink/internal/unix+
-   W    golang.org/x/sys/windows                                     from github.com/tailscale/wireguard-go/conn+
+  LD    golang.org/x/sys/unix                                        from github.com/mdlayher/netlink+
+   W    golang.org/x/sys/windows                                     from github.com/go-ole/go-ole+
    W    golang.org/x/sys/windows/registry                            from golang.zx2c4.com/wireguard/windows/tunnel/winipcfg+
-   W    golang.org/x/sys/windows/svc                                 from tailscale.com/cmd/tailscaled
+   W    golang.org/x/sys/windows/svc                                 from tailscale.com/cmd/tailscaled+
+   W    golang.org/x/sys/windows/svc/mgr                             from tailscale.com/cmd/tailscaled
         golang.org/x/term                                            from tailscale.com/logpolicy
         golang.org/x/text/secure/bidirule                            from golang.org/x/net/idna
         golang.org/x/text/transform                                  from golang.org/x/text/secure/bidirule+
         golang.org/x/text/unicode/bidi                               from golang.org/x/net/idna+
         golang.org/x/text/unicode/norm                               from golang.org/x/net/idna
-        golang.org/x/time/rate                                       from tailscale.com/types/logger+
+        golang.org/x/time/rate                                       from inet.af/netstack/tcpip/stack+
         bufio                                                        from compress/flate+
         bytes                                                        from bufio+
         compress/flate                                               from compress/gzip+
@@ -216,6 +246,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         debug/elf                                                    from rsc.io/goversion/version
         debug/macho                                                  from rsc.io/goversion/version
         debug/pe                                                     from rsc.io/goversion/version
+        embed                                                        from tailscale.com/net/dns+
         encoding                                                     from encoding/json+
         encoding/asn1                                                from crypto/x509+
         encoding/base64                                              from encoding/json+
@@ -223,6 +254,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         encoding/hex                                                 from crypto/x509+
         encoding/json                                                from expvar+
         encoding/pem                                                 from crypto/tls+
+        encoding/xml                                                 from github.com/tailscale/goupnp+
         errors                                                       from bufio+
         expvar                                                       from tailscale.com/derp+
         flag                                                         from tailscale.com/cmd/tailscaled+
@@ -247,7 +279,8 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         net                                                          from crypto/tls+
         net/http                                                     from expvar+
         net/http/httptrace                                           from github.com/tcnksm/go-httpstat+
-        net/http/internal                                            from net/http
+        net/http/httputil                                            from tailscale.com/ipn/localapi
+        net/http/internal                                            from net/http+
         net/http/pprof                                               from tailscale.com/cmd/tailscaled
         net/textproto                                                from golang.org/x/net/http/httpguts+
         net/url                                                      from crypto/x509+

cmd/tailscaled/install_windows.go

@@ -0,0 +1,123 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"time"
+
+	"golang.org/x/sys/windows"
+	"golang.org/x/sys/windows/svc"
+	"golang.org/x/sys/windows/svc/mgr"
+	"tailscale.com/logtail/backoff"
+	"tailscale.com/types/logger"
+	"tailscale.com/util/osshare"
+)
+
+func init() {
+	installSystemDaemon = installSystemDaemonWindows
+	uninstallSystemDaemon = uninstallSystemDaemonWindows
+}
+
+func installSystemDaemonWindows(args []string) (err error) {
+	m, err := mgr.Connect()
+	if err != nil {
+		return fmt.Errorf("failed to connect to Windows service manager: %v", err)
+	}
+
+	service, err := m.OpenService(serviceName)
+	if err == nil {
+		service.Close()
+		return fmt.Errorf("service %q is already installed", serviceName)
+	}
+
+	// no such service; proceed to install the service.
+
+	exe, err := os.Executable()
+	if err != nil {
+		return err
+	}
+
+	c := mgr.Config{
+		ServiceType:  windows.SERVICE_WIN32_OWN_PROCESS,
+		StartType:    mgr.StartAutomatic,
+		ErrorControl: mgr.ErrorNormal,
+		DisplayName:  serviceName,
+		Description:  "Connects this computer to others on the Tailscale network.",
+	}
+
+	service, err = m.CreateService(serviceName, exe, c)
+	if err != nil {
+		return fmt.Errorf("failed to create %q service: %v", serviceName, err)
+	}
+	defer service.Close()
+
+	// Exponential backoff is often too aggressive, so use (mostly)
+	// squares instead.
+	ra := []mgr.RecoveryAction{
+		{mgr.ServiceRestart, 1 * time.Second},
+		{mgr.ServiceRestart, 2 * time.Second},
+		{mgr.ServiceRestart, 4 * time.Second},
+		{mgr.ServiceRestart, 9 * time.Second},
+		{mgr.ServiceRestart, 16 * time.Second},
+		{mgr.ServiceRestart, 25 * time.Second},
+		{mgr.ServiceRestart, 36 * time.Second},
+		{mgr.ServiceRestart, 49 * time.Second},
+		{mgr.ServiceRestart, 64 * time.Second},
+	}
+	const resetPeriodSecs = 60
+	err = service.SetRecoveryActions(ra, resetPeriodSecs)
+	if err != nil {
+		return fmt.Errorf("failed to set service recovery actions: %v", err)
+	}
+
+	return nil
+}
+
+func uninstallSystemDaemonWindows(args []string) (ret error) {
+	// Remove file sharing from Windows shell (noop in non-windows)
+	osshare.SetFileSharingEnabled(false, logger.Discard)
+
+	m, err := mgr.Connect()
+	if err != nil {
+		return fmt.Errorf("failed to connect to Windows service manager: %v", err)
+	}
+	defer m.Disconnect()
+
+	service, err := m.OpenService(serviceName)
+	if err != nil {
+		return fmt.Errorf("failed to open %q service: %v", serviceName, err)
+	}
+
+	st, err := service.Query()
+	if err != nil {
+		service.Close()
+		return fmt.Errorf("failed to query service state: %v", err)
+	}
+	if st.State != svc.Stopped {
+		service.Control(svc.Stop)
+	}
+	err = service.Delete()
+	service.Close()
+	if err != nil {
+		return fmt.Errorf("failed to delete service: %v", err)
+	}
+
+	bo := backoff.NewBackoff("uninstall", logger.Discard, 30*time.Second)
+	end := time.Now().Add(15 * time.Second)
+	for time.Until(end) > 0 {
+		service, err = m.OpenService(serviceName)
+		if err != nil {
+			// service is no longer openable; success!
+			break
+		}
+		service.Close()
+		bo.BackOff(context.Background(), errors.New("service not deleted"))
+	}
+	return nil
+}

cmd/tailscaled/tailscaled.go

@@ -24,38 +24,28 @@ import (
 	"runtime/debug"
 	"strconv"
 	"strings"
-	"sync"
 	"syscall"
 	"time"
 
 	"github.com/go-multierror/multierror"
+	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnserver"
 	"tailscale.com/logpolicy"
 	"tailscale.com/net/dns"
-	"tailscale.com/net/socks5"
+	"tailscale.com/net/socks5/tssocks"
 	"tailscale.com/net/tstun"
 	"tailscale.com/paths"
 	"tailscale.com/types/flagtype"
 	"tailscale.com/types/logger"
-	"tailscale.com/types/netmap"
+	"tailscale.com/util/osshare"
 	"tailscale.com/version"
 	"tailscale.com/version/distro"
 	"tailscale.com/wgengine"
-	"tailscale.com/wgengine/magicsock"
 	"tailscale.com/wgengine/monitor"
 	"tailscale.com/wgengine/netstack"
 	"tailscale.com/wgengine/router"
 )
 
-// globalStateKey is the ipn.StateKey that tailscaled loads on
-// startup.
-//
-// We have to support multiple state keys for other OSes (Windows in
-// particular), but right now Unix daemons run with a single
-// node-global state. To keep open the option of having per-user state
-// later, the global state key doesn't look like a username.
-const globalStateKey = "_daemon"
-
 // defaultTunName returns the default tun device name for the platform.
 func defaultTunName() string {
 	switch runtime.GOOS {
@@ -78,9 +68,13 @@ func defaultTunName() string {
 }
 
 var args struct {
+	// tunname is a /dev/net/tun tunnel name ("tailscale0"), the
+	// string "userspace-networking", "tap:TAPNAME[:BRIDGENAME]"
+	// or comma-separated list thereof.
+	tunname string
+
 	cleanup    bool
 	debug      string
-	tunname    string // tun name, "userspace-networking", or comma-separated list thereof
 	port       uint16
 	statepath  string
 	socketpath string
@@ -114,7 +108,7 @@ func main() {
 	flag.StringVar(&args.debug, "debug", "", "listen address ([ip]:port) of optional debug server")
 	flag.StringVar(&args.socksAddr, "socks5-server", "", `optional [ip]:port to run a SOCK5 server (e.g. "localhost:1080")`)
 	flag.StringVar(&args.tunname, "tun", defaultTunName(), `tunnel interface name; use "userspace-networking" (beta) to not use TUN`)
-	flag.Var(flagtype.PortValue(&args.port, magicsock.DefaultPort), "port", "UDP port to listen on for WireGuard and peer-to-peer traffic; 0 means automatically select")
+	flag.Var(flagtype.PortValue(&args.port, 0), "port", "UDP port to listen on for WireGuard and peer-to-peer traffic; 0 means automatically select")
 	flag.StringVar(&args.statepath, "state", paths.DefaultTailscaledStateFile(), "path of state file")
 	flag.StringVar(&args.socketpath, "socket", paths.DefaultTailscaledSocket(), "path of the service unix socket")
 	flag.BoolVar(&printVersion, "version", false, "print version information and exit")
@@ -148,7 +142,7 @@ func main() {
 		os.Exit(0)
 	}
 
-	if runtime.GOOS == "darwin" && os.Getuid() != 0 && !strings.Contains(args.tunname, "userspace-networking") {
+	if runtime.GOOS == "darwin" && os.Getuid() != 0 && !strings.Contains(args.tunname, "userspace-networking") && !args.cleanup {
 		log.SetFlags(0)
 		log.Fatalf("tailscaled requires root; use sudo tailscaled (or use --tun=userspace-networking)")
 	}
@@ -158,12 +152,39 @@ func main() {
 		log.Fatalf("--socket is required")
 	}
 
-	if err := run(); err != nil {
+	err := run()
+
+	// Remove file sharing from Windows shell (noop in non-windows)
+	osshare.SetFileSharingEnabled(false, logger.Discard)
+
+	if err != nil {
 		// No need to log; the func already did
 		os.Exit(1)
 	}
 }
 
+func ipnServerOpts() (o ipnserver.Options) {
+	// Allow changing the OS-specific IPN behavior for tests
+	// so we can e.g. test Windows-specific behaviors on Linux.
+	goos := os.Getenv("TS_DEBUG_TAILSCALED_IPN_GOOS")
+	if goos == "" {
+		goos = runtime.GOOS
+	}
+
+	o.Port = 41112
+	o.StatePath = args.statepath
+	o.SocketPath = args.socketpath // even for goos=="windows", for tests
+
+	switch goos {
+	default:
+		o.SurviveDisconnects = true
+		o.AutostartStateKey = ipn.GlobalDaemonStateKey
+	case "windows":
+		// Not those.
+	}
+	return o
+}
+
 func run() error {
 	var err error
 
@@ -193,6 +214,9 @@ func run() error {
 	logf = logger.RateLimitedFn(logf, 5*time.Second, 5, 100)
 
 	if args.cleanup {
+		if os.Getenv("TS_PLEASE_PANIC") != "" {
+			panic("TS_PLEASE_PANIC asked us to panic")
+		}
 		dns.Cleanup(logf, args.tunname)
 		router.Cleanup(logf, args.tunname)
 		return nil
@@ -221,6 +245,11 @@ func run() error {
 		if err != nil {
 			log.Fatalf("SOCKS5 listener: %v", err)
 		}
+		if strings.HasSuffix(args.socksAddr, ":0") {
+			// Log kernel-selected port number so integration tests
+			// can find it portably.
+			log.Printf("SOCKS5 listening on %v", socksListener.Addr())
+		}
 	}
 
 	e, useNetstack, err := createEngine(logf, linkMon)
@@ -230,45 +259,13 @@ func run() error {
 	}
 
 	var ns *netstack.Impl
-	if useNetstack {
-		tunDev, magicConn, ok := e.(wgengine.InternalsGetter).GetInternals()
-		if !ok {
-			log.Fatalf("%T is not a wgengine.InternalsGetter", e)
-		}
-		ns, err = netstack.Create(logf, tunDev, e, magicConn)
-		if err != nil {
-			log.Fatalf("netstack.Create: %v", err)
-		}
-		if err := ns.Start(); err != nil {
-			log.Fatalf("failed to start netstack: %v", err)
-		}
+	if useNetstack || wrapNetstack {
+		onlySubnets := wrapNetstack && !useNetstack
+		ns = mustStartNetstack(logf, e, onlySubnets)
 	}
 
 	if socksListener != nil {
-		srv := &socks5.Server{
-			Logf: logger.WithPrefix(logf, "socks5: "),
-		}
-		if useNetstack {
-			srv.Dialer = func(ctx context.Context, network, addr string) (net.Conn, error) {
-				return ns.DialContextTCP(ctx, addr)
-			}
-		} else {
-			var mu sync.Mutex
-			var dns netstack.DNSMap
-			e.AddNetworkMapCallback(func(nm *netmap.NetworkMap) {
-				mu.Lock()
-				defer mu.Unlock()
-				dns = netstack.DNSMapFromNetworkMap(nm)
-			})
-			srv.Dialer = func(ctx context.Context, network, addr string) (net.Conn, error) {
-				ipp, err := dns.Resolve(ctx, addr)
-				if err != nil {
-					return nil, err
-				}
-				var d net.Dialer
-				return d.DialContext(ctx, network, ipp.String())
-			}
-		}
+		srv := tssocks.NewServer(logger.WithPrefix(logf, "socks5: "), e, ns)
 		go func() {
 			log.Fatalf("SOCKS5 server exited: %v", srv.Serve(socksListener))
 		}()
@@ -295,15 +292,8 @@ func run() error {
 		}
 	}()
 
-	opts := ipnserver.Options{
-		SocketPath:         args.socketpath,
-		Port:               41112,
-		StatePath:          args.statepath,
-		AutostartStateKey:  globalStateKey,
-		LegacyConfigPath:   paths.LegacyConfigPath(),
-		SurviveDisconnects: true,
-		DebugMux:           debugMux,
-	}
+	opts := ipnServerOpts()
+	opts.DebugMux = debugMux
 	err = ipnserver.Run(ctx, logf, pol.PublicID.String(), ipnserver.FixedEngine(e), opts)
 	// Cancelation is not an error: it is the only way to stop ipnserver.
 	if err != nil && err != context.Canceled {
@@ -331,6 +321,28 @@ func createEngine(logf logger.Logf, linkMon *monitor.Mon) (e wgengine.Engine, us
 	return nil, false, multierror.New(errs)
 }
 
+var wrapNetstack = shouldWrapNetstack()
+
+func shouldWrapNetstack() bool {
+	if e := os.Getenv("TS_DEBUG_WRAP_NETSTACK"); e != "" {
+		v, err := strconv.ParseBool(e)
+		if err != nil {
+			log.Fatalf("invalid TS_DEBUG_WRAP_NETSTACK value: %v", err)
+		}
+		return v
+	}
+	if distro.Get() == distro.Synology {
+		return true
+	}
+	switch runtime.GOOS {
+	case "windows", "darwin":
+		// Enable on Windows and tailscaled-on-macOS (this doesn't
+		// affect the GUI clients).
+		return true
+	}
+	return false
+}
+
 func tryEngine(logf logger.Logf, linkMon *monitor.Mon, name string) (e wgengine.Engine, useNetstack bool, err error) {
 	conf := wgengine.Config{
 		ListenPort:  args.port,
@@ -338,25 +350,32 @@ func tryEngine(logf logger.Logf, linkMon *monitor.Mon, name string) (e wgengine.
 	}
 	useNetstack = name == "userspace-networking"
 	if !useNetstack {
-		dev, err := tstun.New(logf, name)
+		dev, devName, err := tstun.New(logf, name)
 		if err != nil {
 			tstun.Diagnose(logf, name)
 			return nil, false, err
 		}
 		conf.Tun = dev
-		r, err := router.New(logf, dev)
+		if strings.HasPrefix(name, "tap:") {
+			conf.IsTAP = true
+			e, err := wgengine.NewUserspaceEngine(logf, conf)
+			return e, false, err
+		}
+
+		r, err := router.New(logf, dev, linkMon)
 		if err != nil {
 			dev.Close()
 			return nil, false, err
 		}
+		d, err := dns.NewOSConfigurator(logf, devName)
+		if err != nil {
+			return nil, false, err
+		}
+		conf.DNS = d
 		conf.Router = r
-		tunname, err := dev.Name()
-		if err != nil {
-			r.Close()
-			dev.Close()
-			return nil, false, err
+		if wrapNetstack {
+			conf.Router = netstack.NewSubnetRouterWrapper(conf.Router)
 		}
-		conf.DNS = dns.NewOSConfigurator(logf, tunname)
 	}
 	e, err = wgengine.NewUserspaceEngine(logf, conf)
 	if err != nil {
@@ -384,3 +403,18 @@ func runDebugServer(mux *http.ServeMux, addr string) {
 		log.Fatal(err)
 	}
 }
+
+func mustStartNetstack(logf logger.Logf, e wgengine.Engine, onlySubnets bool) *netstack.Impl {
+	tunDev, magicConn, ok := e.(wgengine.InternalsGetter).GetInternals()
+	if !ok {
+		log.Fatalf("%T is not a wgengine.InternalsGetter", e)
+	}
+	ns, err := netstack.Create(logf, tunDev, e, magicConn, onlySubnets)
+	if err != nil {
+		log.Fatalf("netstack.Create: %v", err)
+	}
+	if err := ns.Start(); err != nil {
+		log.Fatalf("failed to start netstack: %v", err)
+	}
+	return ns
+}

cmd/tailscaled/tailscaled.openrc

@@ -0,0 +1,25 @@
+#!/sbin/openrc-run
+
+set -a
+source /etc/default/tailscaled
+set +a
+
+command="/usr/sbin/tailscaled"
+command_args="--state=/var/lib/tailscale/tailscaled.state --port=$PORT --socket=/var/run/tailscale/tailscaled.sock $FLAGS"
+command_background=true
+pidfile="/run/tailscaled.pid"
+start_stop_daemon_args="-1 /var/log/tailscaled.log -2 /var/log/tailscaled.log"
+
+depend() {
+    need net
+}
+
+start_pre() {
+    mkdir -p /var/run/tailscale
+    mkdir -p /var/lib/tailscale
+    $command --cleanup
+}
+
+stop_post() {
+    $command --cleanup
+}

cmd/tailscaled/tailscaled.service

@@ -2,7 +2,7 @@
 Description=Tailscale node agent
 Documentation=https://tailscale.com/kb/
 Wants=network-pre.target
-After=network-pre.target
+After=network-pre.target NetworkManager.service systemd-resolved.service
 
 [Service]
 EnvironmentFile=/etc/default/tailscaled

cmd/tailscaled/tailscaled_notwindows.go

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build !windows
 // +build !windows
 
 package main // import "tailscale.com/cmd/tailscaled"

cmd/tailscaled/tailscaled_windows.go

@@ -19,22 +19,25 @@ package main // import "tailscale.com/cmd/tailscaled"
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"log"
-	"net"
 	"os"
 	"time"
 
 	"golang.org/x/sys/windows"
 	"golang.org/x/sys/windows/svc"
 	"golang.zx2c4.com/wireguard/windows/tunnel/winipcfg"
+	"inet.af/netaddr"
 	"tailscale.com/ipn/ipnserver"
 	"tailscale.com/logpolicy"
+	"tailscale.com/net/dns"
 	"tailscale.com/net/tstun"
-	"tailscale.com/tempfork/wireguard-windows/firewall"
 	"tailscale.com/types/logger"
 	"tailscale.com/version"
+	"tailscale.com/wf"
 	"tailscale.com/wgengine"
+	"tailscale.com/wgengine/netstack"
 	"tailscale.com/wgengine/router"
 )
 
@@ -125,16 +128,6 @@ func beFirewallKillswitch() bool {
 	log.SetFlags(0)
 	log.Printf("killswitch subprocess starting, tailscale GUID is %s", os.Args[2])
 
-	go func() {
-		b := make([]byte, 16)
-		for {
-			_, err := os.Stdin.Read(b)
-			if err != nil {
-				log.Fatalf("parent process died or requested exit, exiting (%v)", err)
-			}
-		}
-	}()
-
 	guid, err := windows.GUIDFromString(os.Args[2])
 	if err != nil {
 		log.Fatalf("invalid GUID %q: %v", os.Args[2], err)
@@ -142,101 +135,141 @@ func beFirewallKillswitch() bool {
 
 	luid, err := winipcfg.LUIDFromGUID(&guid)
 	if err != nil {
-		log.Fatalf("no interface with GUID %q", guid)
+		log.Fatalf("no interface with GUID %q: %v", guid, err)
 	}
 
-	noProtection := false
-	var dnsIPs []net.IP // unused in called code.
 	start := time.Now()
-	firewall.EnableFirewall(uint64(luid), noProtection, dnsIPs)
+	fw, err := wf.New(uint64(luid))
+	if err != nil {
+		log.Fatalf("failed to enable firewall: %v", err)
+	}
 	log.Printf("killswitch enabled, took %s", time.Since(start))
 
-	// Block until the monitor goroutine shuts us down.
-	select {}
+	// Note(maisem): when local lan access toggled, tailscaled needs to
+	// inform the firewall to let local routes through. The set of routes
+	// is passed in via stdin encoded in json.
+	dcd := json.NewDecoder(os.Stdin)
+	for {
+		var routes []netaddr.IPPrefix
+		if err := dcd.Decode(&routes); err != nil {
+			log.Fatalf("parent process died or requested exit, exiting (%v)", err)
+		}
+		if err := fw.UpdatePermittedRoutes(routes); err != nil {
+			log.Fatalf("failed to update routes (%v)", err)
+		}
+	}
 }
 
 func startIPNServer(ctx context.Context, logid string) error {
 	var logf logger.Logf = log.Printf
-	var eng wgengine.Engine
-	var err error
 
-	getEngine := func() (wgengine.Engine, error) {
-		dev, err := tstun.New(logf, "Tailscale")
+	getEngineRaw := func() (wgengine.Engine, error) {
+		dev, devName, err := tstun.New(logf, "Tailscale")
 		if err != nil {
-			return nil, err
+			return nil, fmt.Errorf("TUN: %w", err)
 		}
-		r, err := router.New(logf, dev)
+		r, err := router.New(logf, dev, nil)
 		if err != nil {
 			dev.Close()
-			return nil, err
+			return nil, fmt.Errorf("router: %w", err)
+		}
+		if wrapNetstack {
+			r = netstack.NewSubnetRouterWrapper(r)
+		}
+		d, err := dns.NewOSConfigurator(logf, devName)
+		if err != nil {
+			r.Close()
+			dev.Close()
+			return nil, fmt.Errorf("DNS: %w", err)
 		}
 		eng, err := wgengine.NewUserspaceEngine(logf, wgengine.Config{
 			Tun:        dev,
 			Router:     r,
+			DNS:        d,
 			ListenPort: 41641,
 		})
 		if err != nil {
 			r.Close()
 			dev.Close()
-			return nil, err
+			return nil, fmt.Errorf("engine: %w", err)
+		}
+		onlySubnets := true
+		if wrapNetstack {
+			mustStartNetstack(logf, eng, onlySubnets)
 		}
 		return wgengine.NewWatchdog(eng), nil
 	}
 
-	if msg := os.Getenv("TS_DEBUG_WIN_FAIL"); msg != "" {
-		err = fmt.Errorf("pretending to be a service failure: %v", msg)
-	} else {
-		// We have a bunch of bug reports of wgengine.NewUserspaceEngine returning a few different errors,
-		// all intermittently. A few times I (Brad) have also seen sporadic failures that simply
-		// restarting fixed. So try a few times.
-		for try := 1; try <= 5; try++ {
-			if try > 1 {
-				// Only sleep a bit. Don't do some massive backoff because
-				// the frontend GUI has a 30 second timeout on connecting to us,
-				// but even 5 seconds is too long for them to get any results.
-				// 5 tries * 1 second each seems fine.
-				time.Sleep(time.Second)
-			}
-			eng, err = getEngine()
+	type engineOrError struct {
+		Engine wgengine.Engine
+		Err    error
+	}
+	engErrc := make(chan engineOrError)
+	t0 := time.Now()
+	go func() {
+		const ms = time.Millisecond
+		for try := 1; ; try++ {
+			logf("tailscaled: getting engine... (try %v)", try)
+			t1 := time.Now()
+			eng, err := getEngineRaw()
+			d, dt := time.Since(t1).Round(ms), time.Since(t1).Round(ms)
 			if err != nil {
-				logf("wgengine.NewUserspaceEngine: (try %v) %v", try, err)
+				logf("tailscaled: engine fetch error (try %v) in %v (total %v, sysUptime %v): %v",
+					try, d, dt, windowsUptime().Round(time.Second), err)
+			} else {
+				if try > 1 {
+					logf("tailscaled: got engine on try %v in %v (total %v)", try, d, dt)
+				} else {
+					logf("tailscaled: got engine in %v", d)
+				}
+			}
+			timer := time.NewTimer(5 * time.Second)
+			engErrc <- engineOrError{eng, err}
+			if err == nil {
+				timer.Stop()
+				return
+			}
+			<-timer.C
+		}
+	}()
+
+	// getEngine is called by ipnserver to get the engine. It's
+	// not called concurrently and is not called again once it
+	// successfully returns an engine.
+	getEngine := func() (wgengine.Engine, error) {
+		if msg := os.Getenv("TS_DEBUG_WIN_FAIL"); msg != "" {
+			return nil, fmt.Errorf("pretending to be a service failure: %v", msg)
+		}
+		for {
+			res := <-engErrc
+			if res.Engine != nil {
+				return res.Engine, nil
+			}
+			if time.Since(t0) < time.Minute || windowsUptime() < 10*time.Minute {
+				// Ignore errors during early boot. Windows 10 auto logs in the GUI
+				// way sooner than the networking stack components start up.
+				// So the network will fail for a bit (and require a few tries) while
+				// the GUI is still fine.
 				continue
 			}
-			if try > 1 {
-				logf("wgengine.NewUserspaceEngine: ended up working on try %v", try)
-			}
-			break
+			// Return nicer errors to users, annotated with logids, which helps
+			// when they file bugs.
+			return nil, fmt.Errorf("%w\n\nlogid: %v", res.Err, logid)
 		}
 	}
-	if err != nil {
-		// Log the error, but don't fatalf. We want to
-		// propagate the error message to the UI frontend. So
-		// we continue and tell the ipnserver to return that
-		// in a Notify message.
-		logf("wgengine.NewUserspaceEngine: %v", err)
-	}
-	opts := ipnserver.Options{
-		Port:               41112,
-		SurviveDisconnects: false,
-		StatePath:          args.statepath,
-	}
-	if err != nil {
-		// Return nicer errors to users, annotated with logids, which helps
-		// when they file bugs.
-		rawGetEngine := getEngine // raw == without verbose logid-containing error
-		getEngine = func() (wgengine.Engine, error) {
-			eng, err := rawGetEngine()
-			if err != nil {
-				return nil, fmt.Errorf("wgengine.NewUserspaceEngine: %v\n\nlogid: %v", err, logid)
-			}
-			return eng, nil
-		}
-	} else {
-		getEngine = ipnserver.FixedEngine(eng)
-	}
-	err = ipnserver.Run(ctx, logf, logid, getEngine, opts)
+	err := ipnserver.Run(ctx, logf, logid, getEngine, ipnServerOpts())
 	if err != nil {
 		logf("ipnserver.Run: %v", err)
 	}
 	return err
 }
+
+var (
+	kernel32           = windows.NewLazySystemDLL("kernel32.dll")
+	getTickCount64Proc = kernel32.NewProc("GetTickCount64")
+)
+
+func windowsUptime() time.Duration {
+	r, _, _ := getTickCount64Proc.Call()
+	return time.Duration(int64(r)) * time.Millisecond
+}

cmd/tsshd/tsshd.go

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build !windows
 // +build !windows
 
 // The tsshd binary is an SSH server that accepts connections
@@ -29,8 +30,8 @@ import (
 	"time"
 	"unsafe"
 
+	"github.com/creack/pty"
 	"github.com/gliderlabs/ssh"
-	"github.com/kr/pty"
 	gossh "golang.org/x/crypto/ssh"
 	"inet.af/netaddr"
 	"tailscale.com/net/interfaces"
@@ -59,11 +60,11 @@ func main() {
 
 	warned := false
 	for {
-		addr, iface, err := interfaces.Tailscale()
+		addrs, iface, err := interfaces.Tailscale()
 		if err != nil {
 			log.Fatalf("listing interfaces: %v", err)
 		}
-		if addr == nil {
+		if len(addrs) == 0 {
 			if !warned {
 				log.Printf("no tailscale interface found; polling until one is available")
 				warned = true
@@ -75,6 +76,13 @@ func main() {
 			continue
 		}
 		warned = false
+		var addr netaddr.IP
+		for _, a := range addrs {
+			if a.Is4() {
+				addr = a
+				break
+			}
+		}
 		listen := net.JoinHostPort(addr.String(), fmt.Sprint(*port))
 		log.Printf("tailscale ssh server listening on %v, %v", iface.Name, listen)
 		s := &ssh.Server{

cmd/tsshd/tsshd_windows.go

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build windows
 // +build windows
 
 package main

control/controlclient/auto.go

@@ -2,18 +2,11 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// Package controlclient implements the client for the Tailscale
-// control plane.
-//
-// It handles authentication, port picking, and collects the local
-// network configuration.
 package controlclient
 
 import (
 	"context"
-	"encoding/json"
 	"fmt"
-	"reflect"
 	"sync"
 	"time"
 
@@ -28,87 +21,28 @@ import (
 	"tailscale.com/types/wgkey"
 )
 
-// State is the high-level state of the client. It is used only in
-// unit tests for proper sequencing, don't depend on it anywhere else.
-// TODO(apenwarr): eliminate 'state', as it's now obsolete.
-type State int
-
-const (
-	StateNew = State(iota)
-	StateNotAuthenticated
-	StateAuthenticating
-	StateURLVisitRequired
-	StateAuthenticated
-	StateSynchronized // connected and received map update
-)
-
-func (s State) MarshalText() ([]byte, error) {
-	return []byte(s.String()), nil
-}
-
-func (s State) String() string {
-	switch s {
-	case StateNew:
-		return "state:new"
-	case StateNotAuthenticated:
-		return "state:not-authenticated"
-	case StateAuthenticating:
-		return "state:authenticating"
-	case StateURLVisitRequired:
-		return "state:url-visit-required"
-	case StateAuthenticated:
-		return "state:authenticated"
-	case StateSynchronized:
-		return "state:synchronized"
-	default:
-		return fmt.Sprintf("state:unknown:%d", int(s))
-	}
-}
-
-type Status struct {
-	_             structs.Incomparable
-	LoginFinished *empty.Message
-	Err           string
-	URL           string
-	Persist       *persist.Persist   // locally persisted configuration
-	NetMap        *netmap.NetworkMap // server-pushed configuration
-	Hostinfo      *tailcfg.Hostinfo  // current Hostinfo data
-	State         State
-}
-
-// Equal reports whether s and s2 are equal.
-func (s *Status) Equal(s2 *Status) bool {
-	if s == nil && s2 == nil {
-		return true
-	}
-	return s != nil && s2 != nil &&
-		(s.LoginFinished == nil) == (s2.LoginFinished == nil) &&
-		s.Err == s2.Err &&
-		s.URL == s2.URL &&
-		reflect.DeepEqual(s.Persist, s2.Persist) &&
-		reflect.DeepEqual(s.NetMap, s2.NetMap) &&
-		reflect.DeepEqual(s.Hostinfo, s2.Hostinfo) &&
-		s.State == s2.State
-}
-
-func (s Status) String() string {
-	b, err := json.MarshalIndent(s, "", "\t")
-	if err != nil {
-		panic(err)
-	}
-	return s.State.String() + " " + string(b)
-}
-
 type LoginGoal struct {
-	_            structs.Incomparable
-	wantLoggedIn bool                 // true if we *want* to be logged in
-	token        *tailcfg.Oauth2Token // oauth token to use when logging in
-	flags        LoginFlags           // flags to use when logging in
-	url          string               // auth url that needs to be visited
+	_               structs.Incomparable
+	wantLoggedIn    bool                 // true if we *want* to be logged in
+	token           *tailcfg.Oauth2Token // oauth token to use when logging in
+	flags           LoginFlags           // flags to use when logging in
+	url             string               // auth url that needs to be visited
+	loggedOutResult chan<- error
 }
 
-// Client connects to a tailcontrol server for a node.
-type Client struct {
+func (g *LoginGoal) sendLogoutError(err error) {
+	if g.loggedOutResult == nil {
+		return
+	}
+	select {
+	case g.loggedOutResult <- err:
+	default:
+	}
+}
+
+// Auto connects to a tailcontrol server for a node.
+// It's a concrete implementation of the Client interface.
+type Auto struct {
 	direct   *Direct // our interface to the server APIs
 	timeNow  func() time.Time
 	logf     logger.Logf
@@ -141,8 +75,8 @@ type Client struct {
 	mapDone    chan struct{}   // when closed, map goroutine is done
 }
 
-// New creates and starts a new Client.
-func New(opts Options) (*Client, error) {
+// New creates and starts a new Auto.
+func New(opts Options) (*Auto, error) {
 	c, err := NewNoStart(opts)
 	if c != nil {
 		c.Start()
@@ -150,8 +84,8 @@ func New(opts Options) (*Client, error) {
 	return c, err
 }
 
-// NewNoStart creates a new Client, but without calling Start on it.
-func NewNoStart(opts Options) (*Client, error) {
+// NewNoStart creates a new Auto, but without calling Start on it.
+func NewNoStart(opts Options) (*Auto, error) {
 	direct, err := NewDirect(opts)
 	if err != nil {
 		return nil, err
@@ -162,7 +96,7 @@ func NewNoStart(opts Options) (*Client, error) {
 	if opts.TimeNow == nil {
 		opts.TimeNow = time.Now
 	}
-	c := &Client{
+	c := &Auto{
 		direct:   direct,
 		timeNow:  opts.TimeNow,
 		logf:     opts.Logf,
@@ -178,7 +112,7 @@ func NewNoStart(opts Options) (*Client, error) {
 
 }
 
-func (c *Client) onHealthChange(sys health.Subsystem, err error) {
+func (c *Auto) onHealthChange(sys health.Subsystem, err error) {
 	if sys == health.SysOverall {
 		return
 	}
@@ -189,12 +123,13 @@ func (c *Client) onHealthChange(sys health.Subsystem, err error) {
 // SetPaused controls whether HTTP activity should be paused.
 //
 // The client can be paused and unpaused repeatedly, unlike Start and Shutdown, which can only be used once.
-func (c *Client) SetPaused(paused bool) {
+func (c *Auto) SetPaused(paused bool) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	if paused == c.paused {
 		return
 	}
+	c.logf("setPaused(%v)", paused)
 	c.paused = paused
 	if paused {
 		// Only cancel the map routine. (The auth routine isn't expensive
@@ -211,7 +146,7 @@ func (c *Client) SetPaused(paused bool) {
 // Start starts the client's goroutines.
 //
 // It should only be called for clients created by NewNoStart.
-func (c *Client) Start() {
+func (c *Auto) Start() {
 	go c.authRoutine()
 	go c.mapRoutine()
 }
@@ -221,7 +156,7 @@ func (c *Client) Start() {
 // streaming response open), or start a new streaming one if necessary.
 //
 // It should be called whenever there's something new to tell the server.
-func (c *Client) sendNewMapRequest() {
+func (c *Auto) sendNewMapRequest() {
 	c.mu.Lock()
 
 	// If we're not already streaming a netmap, or if we're already stuck
@@ -260,7 +195,7 @@ func (c *Client) sendNewMapRequest() {
 	}()
 }
 
-func (c *Client) cancelAuth() {
+func (c *Auto) cancelAuth() {
 	c.mu.Lock()
 	if c.authCancel != nil {
 		c.authCancel()
@@ -271,7 +206,7 @@ func (c *Client) cancelAuth() {
 	c.mu.Unlock()
 }
 
-func (c *Client) cancelMapLocked() {
+func (c *Auto) cancelMapLocked() {
 	if c.mapCancel != nil {
 		c.mapCancel()
 	}
@@ -280,13 +215,13 @@ func (c *Client) cancelMapLocked() {
 	}
 }
 
-func (c *Client) cancelMapUnsafely() {
+func (c *Auto) cancelMapUnsafely() {
 	c.mu.Lock()
 	c.cancelMapLocked()
 	c.mu.Unlock()
 }
 
-func (c *Client) cancelMapSafely() {
+func (c *Auto) cancelMapSafely() {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
@@ -322,7 +257,7 @@ func (c *Client) cancelMapSafely() {
 	}
 }
 
-func (c *Client) authRoutine() {
+func (c *Auto) authRoutine() {
 	defer close(c.authDone)
 	bo := backoff.NewBackoff("authRoutine", c.logf, 30*time.Second)
 
@@ -333,7 +268,7 @@ func (c *Client) authRoutine() {
 		if goal != nil {
 			c.logf("authRoutine: %s; wantLoggedIn=%v", c.state, goal.wantLoggedIn)
 		} else {
-			c.logf("authRoutine: %s; goal=nil", c.state)
+			c.logf("authRoutine: %s; goal=nil paused=%v", c.state, c.paused)
 		}
 		c.mu.Unlock()
 
@@ -363,6 +298,7 @@ func (c *Client) authRoutine() {
 
 		if !goal.wantLoggedIn {
 			err := c.direct.TryLogout(ctx)
+			goal.sendLogoutError(err)
 			if err != nil {
 				report(err, "TryLogout")
 				bo.BackOff(ctx, err)
@@ -402,7 +338,8 @@ func (c *Client) authRoutine() {
 				report(err, f)
 				bo.BackOff(ctx, err)
 				continue
-			} else if url != "" {
+			}
+			if url != "" {
 				if goal.url != "" {
 					err = fmt.Errorf("[unexpected] server required a new URL?")
 					report(err, "WaitLoginURL")
@@ -439,7 +376,7 @@ func (c *Client) authRoutine() {
 
 // Expiry returns the credential expiration time, or the zero time if
 // the expiration time isn't known. Used in tests only.
-func (c *Client) Expiry() *time.Time {
+func (c *Auto) Expiry() *time.Time {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	return c.expiry
@@ -447,21 +384,21 @@ func (c *Client) Expiry() *time.Time {
 
 // Direct returns the underlying direct client object. Used in tests
 // only.
-func (c *Client) Direct() *Direct {
+func (c *Auto) Direct() *Direct {
 	return c.direct
 }
 
 // unpausedChanLocked returns a new channel that is closed when the
-// current Client pause is unpaused.
+// current Auto pause is unpaused.
 //
 // c.mu must be held
-func (c *Client) unpausedChanLocked() <-chan struct{} {
+func (c *Auto) unpausedChanLocked() <-chan struct{} {
 	unpaused := make(chan struct{})
 	c.unpauseWaiters = append(c.unpauseWaiters, unpaused)
 	return unpaused
 }
 
-func (c *Client) mapRoutine() {
+func (c *Auto) mapRoutine() {
 	defer close(c.mapDone)
 	bo := backoff.NewBackoff("mapRoutine", c.logf, 30*time.Second)
 
@@ -583,7 +520,10 @@ func (c *Client) mapRoutine() {
 	}
 }
 
-func (c *Client) AuthCantContinue() bool {
+func (c *Auto) AuthCantContinue() bool {
+	if c == nil {
+		return true
+	}
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
@@ -591,13 +531,13 @@ func (c *Client) AuthCantContinue() bool {
 }
 
 // SetStatusFunc sets fn as the callback to run on any status change.
-func (c *Client) SetStatusFunc(fn func(Status)) {
+func (c *Auto) SetStatusFunc(fn func(Status)) {
 	c.mu.Lock()
 	c.statusFunc = fn
 	c.mu.Unlock()
 }
 
-func (c *Client) SetHostinfo(hi *tailcfg.Hostinfo) {
+func (c *Auto) SetHostinfo(hi *tailcfg.Hostinfo) {
 	if hi == nil {
 		panic("nil Hostinfo")
 	}
@@ -610,7 +550,7 @@ func (c *Client) SetHostinfo(hi *tailcfg.Hostinfo) {
 	c.sendNewMapRequest()
 }
 
-func (c *Client) SetNetInfo(ni *tailcfg.NetInfo) {
+func (c *Auto) SetNetInfo(ni *tailcfg.NetInfo) {
 	if ni == nil {
 		panic("nil NetInfo")
 	}
@@ -623,7 +563,7 @@ func (c *Client) SetNetInfo(ni *tailcfg.NetInfo) {
 	c.sendNewMapRequest()
 }
 
-func (c *Client) sendStatus(who string, err error, url string, nm *netmap.NetworkMap) {
+func (c *Auto) sendStatus(who string, err error, url string, nm *netmap.NetworkMap) {
 	c.mu.Lock()
 	state := c.state
 	loggedIn := c.loggedIn
@@ -636,9 +576,12 @@ func (c *Client) sendStatus(who string, err error, url string, nm *netmap.Networ
 	c.logf("[v1] sendStatus: %s: %v", who, state)
 
 	var p *persist.Persist
-	var fin *empty.Message
+	var loginFin, logoutFin *empty.Message
 	if state == StateAuthenticated {
-		fin = new(empty.Message)
+		loginFin = new(empty.Message)
+	}
+	if state == StateNotAuthenticated {
+		logoutFin = new(empty.Message)
 	}
 	if nm != nil && loggedIn && synced {
 		pp := c.direct.GetPersist()
@@ -649,12 +592,13 @@ func (c *Client) sendStatus(who string, err error, url string, nm *netmap.Networ
 		nm = nil
 	}
 	new := Status{
-		LoginFinished: fin,
-		URL:           url,
-		Persist:       p,
-		NetMap:        nm,
-		Hostinfo:      hi,
-		State:         state,
+		LoginFinished:  loginFin,
+		LogoutFinished: logoutFin,
+		URL:            url,
+		Persist:        p,
+		NetMap:         nm,
+		Hostinfo:       hi,
+		State:          state,
 	}
 	if err != nil {
 		new.Err = err.Error()
@@ -668,7 +612,7 @@ func (c *Client) sendStatus(who string, err error, url string, nm *netmap.Networ
 	c.mu.Unlock()
 }
 
-func (c *Client) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
+func (c *Auto) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
 	c.logf("client.Login(%v, %v)", t != nil, flags)
 
 	c.mu.Lock()
@@ -682,18 +626,42 @@ func (c *Client) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
 	c.cancelAuth()
 }
 
-func (c *Client) Logout() {
-	c.logf("client.Logout()")
+func (c *Auto) StartLogout() {
+	c.logf("client.StartLogout()")
 
 	c.mu.Lock()
 	c.loginGoal = &LoginGoal{
 		wantLoggedIn: false,
 	}
 	c.mu.Unlock()
-
 	c.cancelAuth()
 }
 
+func (c *Auto) Logout(ctx context.Context) error {
+	c.logf("client.Logout()")
+
+	errc := make(chan error, 1)
+
+	c.mu.Lock()
+	c.loginGoal = &LoginGoal{
+		wantLoggedIn:    false,
+		loggedOutResult: errc,
+	}
+	c.mu.Unlock()
+	c.cancelAuth()
+
+	timer := time.NewTimer(10 * time.Second)
+	defer timer.Stop()
+	select {
+	case err := <-errc:
+		return err
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		return context.DeadlineExceeded
+	}
+}
+
 // UpdateEndpoints sets the client's discovered endpoints and sends
 // them to the control server if they've changed.
 //
@@ -701,14 +669,14 @@ func (c *Client) Logout() {
 //
 // The localPort field is unused except for integration tests in
 // another repo.
-func (c *Client) UpdateEndpoints(localPort uint16, endpoints []string) {
+func (c *Auto) UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) {
 	changed := c.direct.SetEndpoints(localPort, endpoints)
 	if changed {
 		c.sendNewMapRequest()
 	}
 }
 
-func (c *Client) Shutdown() {
+func (c *Auto) Shutdown() {
 	c.logf("client.Shutdown()")
 
 	c.mu.Lock()
@@ -734,17 +702,23 @@ func (c *Client) Shutdown() {
 
 // NodePublicKey returns the node public key currently in use. This is
 // used exclusively in tests.
-func (c *Client) TestOnlyNodePublicKey() wgkey.Key {
+func (c *Auto) TestOnlyNodePublicKey() wgkey.Key {
 	priv := c.direct.GetPersist()
 	return priv.PrivateNodeKey.Public()
 }
 
-func (c *Client) TestOnlySetAuthKey(authkey string) {
+func (c *Auto) TestOnlySetAuthKey(authkey string) {
 	c.direct.mu.Lock()
 	defer c.direct.mu.Unlock()
 	c.direct.authKey = authkey
 }
 
-func (c *Client) TestOnlyTimeNow() time.Time {
+func (c *Auto) TestOnlyTimeNow() time.Time {
 	return c.timeNow()
 }
+
+// SetDNS sends the SetDNSRequest request to the control plane server,
+// requesting a DNS record be created or updated.
+func (c *Auto) SetDNS(ctx context.Context, req *tailcfg.SetDNSRequest) error {
+	return c.direct.SetDNS(ctx, req)
+}

control/controlclient/client.go

@@ -0,0 +1,80 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package controlclient implements the client for the Tailscale
+// control plane.
+//
+// It handles authentication, port picking, and collects the local
+// network configuration.
+package controlclient
+
+import (
+	"context"
+
+	"tailscale.com/tailcfg"
+)
+
+type LoginFlags int
+
+const (
+	LoginDefault     = LoginFlags(0)
+	LoginInteractive = LoginFlags(1 << iota) // force user login and key refresh
+)
+
+// Client represents a client connection to the control server.
+// Currently this is done through a pair of polling https requests in
+// the Auto client, but that might change eventually.
+type Client interface {
+	// SetStatusFunc provides a callback to call when control sends us
+	// a message.
+	SetStatusFunc(func(Status))
+	// Shutdown closes this session, which should not be used any further
+	// afterwards.
+	Shutdown()
+	// Login begins an interactive or non-interactive login process.
+	// Client will eventually call the Status callback with either a
+	// LoginFinished flag (on success) or an auth URL (if further
+	// interaction is needed).
+	Login(*tailcfg.Oauth2Token, LoginFlags)
+	// StartLogout starts an asynchronous logout process.
+	// When it finishes, the Status callback will be called while
+	// AuthCantContinue()==true.
+	StartLogout()
+	// Logout starts a synchronous logout process. It doesn't return
+	// until the logout operation has been completed.
+	Logout(context.Context) error
+	// SetPaused pauses or unpauses the controlclient activity as much
+	// as possible, without losing its internal state, to minimize
+	// unnecessary network activity.
+	// TODO: It might be better to simply shutdown the controlclient and
+	// make a new one when it's time to unpause.
+	SetPaused(bool)
+	// AuthCantContinue returns whether authentication is blocked. If it
+	// is, you either need to visit the auth URL (previously sent in a
+	// Status callback) or call the Login function appropriately.
+	// TODO: this probably belongs in the Status itself instead.
+	AuthCantContinue() bool
+	// SetHostinfo changes the Hostinfo structure that will be sent in
+	// subsequent node registration requests.
+	// TODO: a server-side change would let us simply upload this
+	// in a separate http request. It has nothing to do with the rest of
+	// the state machine.
+	SetHostinfo(*tailcfg.Hostinfo)
+	// SetNetinfo changes the NetIinfo structure that will be sent in
+	// subsequent node registration requests.
+	// TODO: a server-side change would let us simply upload this
+	// in a separate http request. It has nothing to do with the rest of
+	// the state machine.
+	SetNetInfo(*tailcfg.NetInfo)
+	// UpdateEndpoints changes the Endpoint structure that will be sent
+	// in subsequent node registration requests.
+	// TODO: localPort seems to be obsolete, remove it.
+	// TODO: a server-side change would let us simply upload this
+	// in a separate http request. It has nothing to do with the rest of
+	// the state machine.
+	UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint)
+	// SetDNS sends the SetDNSRequest request to the control plane server,
+	// requesting a DNS record be created or updated.
+	SetDNS(context.Context, *tailcfg.SetDNSRequest) error
+}

control/controlclient/controlclient_test.go

@@ -22,7 +22,7 @@ func fieldsOf(t reflect.Type) (fields []string) {
 
 func TestStatusEqual(t *testing.T) {
 	// Verify that the Equal method stays in sync with reality
-	equalHandles := []string{"LoginFinished", "Err", "URL", "Persist", "NetMap", "Hostinfo", "State"}
+	equalHandles := []string{"LoginFinished", "LogoutFinished", "Err", "URL", "NetMap", "State", "Persist", "Hostinfo"}
 	if have := fieldsOf(reflect.TypeOf(Status{})); !reflect.DeepEqual(have, equalHandles) {
 		t.Errorf("Status.Equal check might be out of sync\nfields: %q\nhandled: %q\n",
 			have, equalHandles)

control/controlclient/debug.go

@@ -11,7 +11,6 @@ import (
 	"fmt"
 	"log"
 	"net/http"
-	"regexp"
 	"runtime"
 	"strconv"
 	"time"
@@ -42,28 +41,82 @@ func dumpGoroutinesToURL(c *http.Client, targetURL string) {
 	}
 }
 
-var reHexArgs = regexp.MustCompile(`\b0x[0-9a-f]+\b`)
-
 // scrubbedGoroutineDump returns the list of all current goroutines, but with the actual
 // values of arguments scrubbed out, lest it contain some private key material.
 func scrubbedGoroutineDump() []byte {
-	buf := make([]byte, 1<<20)
-	buf = buf[:runtime.Stack(buf, true)]
+	var buf []byte
+	// Grab stacks multiple times into increasingly larger buffer sizes
+	// to minimize the risk that we blow past our iOS memory limit.
+	for size := 1 << 10; size <= 1<<20; size += 1 << 10 {
+		buf = make([]byte, size)
+		buf = buf[:runtime.Stack(buf, true)]
+		if len(buf) < size {
+			// It fit.
+			break
+		}
+	}
+	return scrubHex(buf)
+}
 
+func scrubHex(buf []byte) []byte {
 	saw := map[string][]byte{} // "0x123" => "v1%3" (unique value 1 and its value mod 8)
-	return reHexArgs.ReplaceAllFunc(buf, func(in []byte) []byte {
+
+	foreachHexAddress(buf, func(in []byte) {
 		if string(in) == "0x0" {
-			return in
+			return
 		}
 		if v, ok := saw[string(in)]; ok {
-			return v
+			for i := range in {
+				in[i] = '_'
+			}
+			copy(in, v)
+			return
 		}
+		inStr := string(in)
 		u64, err := strconv.ParseUint(string(in[2:]), 16, 64)
+		for i := range in {
+			in[i] = '_'
+		}
 		if err != nil {
-			return []byte("??")
+			in[0] = '?'
+			return
 		}
 		v := []byte(fmt.Sprintf("v%d%%%d", len(saw)+1, u64%8))
-		saw[string(in)] = v
-		return v
+		saw[inStr] = v
+		copy(in, v)
 	})
+	return buf
+}
+
+var ohx = []byte("0x")
+
+// foreachHexAddress calls f with each subslice of b that matches
+// regexp `0x[0-9a-f]*`.
+func foreachHexAddress(b []byte, f func([]byte)) {
+	for len(b) > 0 {
+		i := bytes.Index(b, ohx)
+		if i == -1 {
+			return
+		}
+		b = b[i:]
+		hx := hexPrefix(b)
+		f(hx)
+		b = b[len(hx):]
+	}
+}
+
+func hexPrefix(b []byte) []byte {
+	for i, c := range b {
+		if i < 2 {
+			continue
+		}
+		if !isHexByte(c) {
+			return b[:i]
+		}
+	}
+	return b
+}
+
+func isHexByte(b byte) bool {
+	return '0' <= b && b <= '9' || 'a' <= b && b <= 'f' || 'A' <= b && b <= 'F'
 }

control/controlclient/debug_test.go

@@ -9,3 +9,22 @@ import "testing"
 func TestScrubbedGoroutineDump(t *testing.T) {
 	t.Logf("Got:\n%s\n", scrubbedGoroutineDump())
 }
+
+func TestScrubHex(t *testing.T) {
+	tests := []struct {
+		in, want string
+	}{
+		{"foo", "foo"},
+		{"", ""},
+		{"0x", "?_"},
+		{"0x001 and same 0x001", "v1%1_ and same v1%1_"},
+		{"0x008 and same 0x008", "v1%0_ and same v1%0_"},
+		{"0x001 and diff 0x002", "v1%1_ and diff v2%2_"},
+	}
+	for _, tt := range tests {
+		got := scrubHex([]byte(tt.in))
+		if string(got) != tt.want {
+			t.Errorf("for input:\n%s\n\ngot:\n%s\n\nwant:\n%s\n", tt.in, got, tt.want)
+		}
+	}
+}

control/controlclient/direct.go

@@ -23,7 +23,6 @@ import (
 	"path/filepath"
 	"reflect"
 	"runtime"
-	"sort"
 	"strconv"
 	"strings"
 	"sync"
@@ -32,7 +31,9 @@ import (
 
 	"golang.org/x/crypto/nacl/box"
 	"inet.af/netaddr"
+	"tailscale.com/control/controlknobs"
 	"tailscale.com/health"
+	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/log/logheap"
 	"tailscale.com/net/dnscache"
 	"tailscale.com/net/dnsfallback"
@@ -46,9 +47,9 @@ import (
 	"tailscale.com/types/opt"
 	"tailscale.com/types/persist"
 	"tailscale.com/types/wgkey"
+	"tailscale.com/util/dnsname"
 	"tailscale.com/util/systemd"
 	"tailscale.com/version"
-	"tailscale.com/wgengine/filter"
 	"tailscale.com/wgengine/monitor"
 )
 
@@ -67,6 +68,7 @@ type Direct struct {
 	debugFlags             []string
 	keepSharerAndUserSplit bool
 	skipIPForwardingCheck  bool
+	pinger                 Pinger
 
 	mu           sync.Mutex // mutex guards the following fields
 	serverKey    wgkey.Key
@@ -76,9 +78,10 @@ type Direct struct {
 	expiry       *time.Time
 	// hostinfo is mutated in-place while mu is held.
 	hostinfo      *tailcfg.Hostinfo // always non-nil
-	endpoints     []string
+	endpoints     []tailcfg.Endpoint
 	everEndpoints bool   // whether we've ever had non-empty endpoints
 	localPort     uint16 // or zero to mean auto
+	lastPingURL   string // last PingRequest.URL received, for dup suppresion
 }
 
 type Options struct {
@@ -104,6 +107,18 @@ type Options struct {
 	// forwarding works and should not be double-checked by the
 	// controlclient package.
 	SkipIPForwardingCheck bool
+
+	// Pinger optionally specifies the Pinger to use to satisfy
+	// MapResponse.PingRequest queries from the control plane.
+	// If nil, PingRequest queries are not answered.
+	Pinger Pinger
+}
+
+// Pinger is a subset of the wgengine.Engine interface, containing just the Ping method.
+type Pinger interface {
+	// Ping is a request to start a discovery or TSMP ping with the peer handling
+	// the given IP and then call cb with its ping latency & method.
+	Ping(ip netaddr.IP, useTSMP bool, cb func(*ipnstate.PingResult))
 }
 
 type Decompressor interface {
@@ -144,7 +159,7 @@ func NewDirect(opts Options) (*Direct, error) {
 		tr := http.DefaultTransport.(*http.Transport).Clone()
 		tr.Proxy = tshttpproxy.ProxyFromEnvironment
 		tshttpproxy.SetTransportGetProxyConnectHeader(tr)
-		tr.TLSClientConfig = tlsdial.Config(serverURL.Host, tr.TLSClientConfig)
+		tr.TLSClientConfig = tlsdial.Config(serverURL.Hostname(), tr.TLSClientConfig)
 		tr.DialContext = dnscache.Dialer(dialer.DialContext, dnsCache)
 		tr.DialTLSContext = dnscache.TLSDialer(dialer.DialContext, dnsCache, tr.TLSClientConfig)
 		tr.ForceAttemptHTTP2 = true
@@ -166,6 +181,7 @@ func NewDirect(opts Options) (*Direct, error) {
 		keepSharerAndUserSplit: opts.KeepSharerAndUserSplit,
 		linkMon:                opts.LinkMonitor,
 		skipIPForwardingCheck:  opts.SkipIPForwardingCheck,
+		pinger:                 opts.Pinger,
 	}
 	if opts.Hostinfo == nil {
 		c.SetHostinfo(NewHostinfo())
@@ -179,6 +195,7 @@ var osVersion func() string // non-nil on some platforms
 
 func NewHostinfo() *tailcfg.Hostinfo {
 	hostname, _ := os.Hostname()
+	hostname = dnsname.FirstLabel(hostname)
 	var osv string
 	if osVersion != nil {
 		osv = osVersion()
@@ -203,6 +220,13 @@ func packageType() string {
 		// Using tailscaled or IPNExtension?
 		exe, _ := os.Executable()
 		return filepath.Base(exe)
+	case "linux":
+		// Report whether this is in a snap.
+		// See https://snapcraft.io/docs/environment-variables
+		// We just look at two somewhat arbitrarily.
+		if os.Getenv("SNAP_NAME") != "" && os.Getenv("SNAP") != "" {
+			return "snap"
+		}
 	}
 	return ""
 }
@@ -251,30 +275,22 @@ func (c *Direct) GetPersist() persist.Persist {
 	return c.persist
 }
 
-type LoginFlags int
-
-const (
-	LoginDefault     = LoginFlags(0)
-	LoginInteractive = LoginFlags(1 << iota) // force user login and key refresh
-)
-
 func (c *Direct) TryLogout(ctx context.Context) error {
 	c.logf("direct.TryLogout()")
 
-	c.mu.Lock()
-	defer c.mu.Unlock()
+	mustRegen, newURL, err := c.doLogin(ctx, loginOpt{Logout: true})
+	c.logf("TryLogout control response: mustRegen=%v, newURL=%v, err=%v", mustRegen, newURL, err)
 
-	// TODO(crawshaw): Tell the server. This node key should be
-	// immediately invalidated.
-	//if !c.persist.PrivateNodeKey.IsZero() {
-	//}
+	c.mu.Lock()
 	c.persist = persist.Persist{}
-	return nil
+	c.mu.Unlock()
+
+	return err
 }
 
 func (c *Direct) TryLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags LoginFlags) (url string, err error) {
 	c.logf("direct.TryLogin(token=%v, flags=%v)", t != nil, flags)
-	return c.doLoginOrRegen(ctx, t, flags, false, "")
+	return c.doLoginOrRegen(ctx, loginOpt{Token: t, Flags: flags})
 }
 
 // WaitLoginURL sits in a long poll waiting for the user to authenticate at url.
@@ -282,22 +298,30 @@ func (c *Direct) TryLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Log
 // On success, newURL and err will both be nil.
 func (c *Direct) WaitLoginURL(ctx context.Context, url string) (newURL string, err error) {
 	c.logf("direct.WaitLoginURL")
-	return c.doLoginOrRegen(ctx, nil, LoginDefault, false, url)
+	return c.doLoginOrRegen(ctx, loginOpt{URL: url})
 }
 
-func (c *Direct) doLoginOrRegen(ctx context.Context, t *tailcfg.Oauth2Token, flags LoginFlags, regen bool, url string) (newURL string, err error) {
-	mustregen, url, err := c.doLogin(ctx, t, flags, regen, url)
+func (c *Direct) doLoginOrRegen(ctx context.Context, opt loginOpt) (newURL string, err error) {
+	mustRegen, url, err := c.doLogin(ctx, opt)
 	if err != nil {
 		return url, err
 	}
-	if mustregen {
-		_, url, err = c.doLogin(ctx, t, flags, true, url)
+	if mustRegen {
+		opt.Regen = true
+		_, url, err = c.doLogin(ctx, opt)
 	}
-
 	return url, err
 }
 
-func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags LoginFlags, regen bool, url string) (mustregen bool, newurl string, err error) {
+type loginOpt struct {
+	Token  *tailcfg.Oauth2Token
+	Flags  LoginFlags
+	Regen  bool
+	URL    string
+	Logout bool
+}
+
+func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, newURL string, err error) {
 	c.mu.Lock()
 	persist := c.persist
 	tryingNewKey := c.tryingNewKey
@@ -316,23 +340,29 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 		return false, "", errors.New("getMachinePrivKey returned zero key")
 	}
 
-	if expired {
-		c.logf("Old key expired -> regen=true")
-		systemd.Status("key expired; run 'tailscale up' to authenticate")
-		regen = true
-	}
-	if (flags & LoginInteractive) != 0 {
-		c.logf("LoginInteractive -> regen=true")
-		regen = true
+	regen := opt.Regen
+	if opt.Logout {
+		c.logf("logging out...")
+	} else {
+		if expired {
+			c.logf("Old key expired -> regen=true")
+			systemd.Status("key expired; run 'tailscale up' to authenticate")
+			regen = true
+		}
+		if (opt.Flags & LoginInteractive) != 0 {
+			c.logf("LoginInteractive -> regen=true")
+			regen = true
+		}
 	}
 
-	c.logf("doLogin(regen=%v, hasUrl=%v)", regen, url != "")
+	c.logf("doLogin(regen=%v, hasUrl=%v)", regen, opt.URL != "")
 	if serverKey.IsZero() {
 		var err error
 		serverKey, err = loadServerKey(ctx, c.httpc, c.serverURL)
 		if err != nil {
-			return regen, url, err
+			return regen, opt.URL, err
 		}
+		c.logf("control server key %s from %s", serverKey.HexString(), c.serverURL)
 
 		c.mu.Lock()
 		c.serverKey = serverKey
@@ -340,17 +370,21 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 	}
 
 	var oldNodeKey wgkey.Key
-	if url != "" {
-	} else if regen || persist.PrivateNodeKey.IsZero() {
+	switch {
+	case opt.Logout:
+		tryingNewKey = persist.PrivateNodeKey
+	case opt.URL != "":
+		// Nothing.
+	case regen || persist.PrivateNodeKey.IsZero():
 		c.logf("Generating a new nodekey.")
 		persist.OldPrivateNodeKey = persist.PrivateNodeKey
 		key, err := wgkey.NewPrivate()
 		if err != nil {
 			c.logf("login keygen: %v", err)
-			return regen, url, err
+			return regen, opt.URL, err
 		}
 		tryingNewKey = key
-	} else {
+	default:
 		// Try refreshing the current key first
 		tryingNewKey = persist.PrivateNodeKey
 	}
@@ -359,11 +393,14 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 	}
 
 	if tryingNewKey.IsZero() {
+		if opt.Logout {
+			return false, "", errors.New("no nodekey to log out")
+		}
 		log.Fatalf("tryingNewKey is empty, give up")
 	}
 	if backendLogID == "" {
 		err = errors.New("hostinfo: BackendLogID missing")
-		return regen, url, err
+		return regen, opt.URL, err
 	}
 	now := time.Now().Round(time.Second)
 	request := tailcfg.RegisterRequest{
@@ -371,13 +408,16 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 		OldNodeKey: tailcfg.NodeKey(oldNodeKey),
 		NodeKey:    tailcfg.NodeKey(tryingNewKey.Public()),
 		Hostinfo:   hostinfo,
-		Followup:   url,
+		Followup:   opt.URL,
 		Timestamp:  &now,
 	}
+	if opt.Logout {
+		request.Expiry = time.Unix(123, 0) // far in the past
+	}
 	c.logf("RegisterReq: onode=%v node=%v fup=%v",
 		request.OldNodeKey.ShortString(),
-		request.NodeKey.ShortString(), url != "")
-	request.Auth.Oauth2Token = t
+		request.NodeKey.ShortString(), opt.URL != "")
+	request.Auth.Oauth2Token = opt.Token
 	request.Auth.Provider = persist.Provider
 	request.Auth.LoginName = persist.LoginName
 	request.Auth.AuthKey = authKey
@@ -391,38 +431,48 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 
 		// Don't log the common error types. Signatures are not usually enabled,
 		// so these are expected.
-		if err != errCertificateNotConfigured && err != errNoCertStore {
+		if !errors.Is(err, errCertificateNotConfigured) && !errors.Is(err, errNoCertStore) {
 			c.logf("RegisterReq sign error: %v", err)
 		}
 	}
+	if debugRegister {
+		j, _ := json.MarshalIndent(request, "", "\t")
+		c.logf("RegisterRequest: %s", j)
+	}
+
 	bodyData, err := encode(request, &serverKey, &machinePrivKey)
 	if err != nil {
-		return regen, url, err
+		return regen, opt.URL, err
 	}
 	body := bytes.NewReader(bodyData)
 
 	u := fmt.Sprintf("%s/machine/%s", c.serverURL, machinePrivKey.Public().HexString())
 	req, err := http.NewRequest("POST", u, body)
 	if err != nil {
-		return regen, url, err
+		return regen, opt.URL, err
 	}
 	req = req.WithContext(ctx)
 
 	res, err := c.httpc.Do(req)
 	if err != nil {
-		return regen, url, fmt.Errorf("register request: %v", err)
+		return regen, opt.URL, fmt.Errorf("register request: %v", err)
 	}
 	if res.StatusCode != 200 {
 		msg, _ := ioutil.ReadAll(res.Body)
 		res.Body.Close()
-		return regen, url, fmt.Errorf("register request: http %d: %.200s",
+		return regen, opt.URL, fmt.Errorf("register request: http %d: %.200s",
 			res.StatusCode, strings.TrimSpace(string(msg)))
 	}
 	resp := tailcfg.RegisterResponse{}
 	if err := decode(res, &resp, &serverKey, &machinePrivKey); err != nil {
 		c.logf("error decoding RegisterResponse with server key %s and machine key %s: %v", serverKey, machinePrivKey.Public(), err)
-		return regen, url, fmt.Errorf("register request: %v", err)
+		return regen, opt.URL, fmt.Errorf("register request: %v", err)
 	}
+	if debugRegister {
+		j, _ := json.MarshalIndent(resp, "", "\t")
+		c.logf("RegisterResponse: %s", j)
+	}
+
 	// Log without PII:
 	c.logf("RegisterReq: got response; nodeKeyExpired=%v, machineAuthorized=%v; authURL=%v",
 		resp.NodeKeyExpired, resp.MachineAuthorized, resp.AuthURL != "")
@@ -435,10 +485,10 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 			request.NodeKey.ShortString())
 		return true, "", nil
 	}
-	if persist.Provider == "" {
+	if resp.Login.Provider != "" {
 		persist.Provider = resp.Login.Provider
 	}
-	if persist.LoginName == "" {
+	if resp.Login.LoginName != "" {
 		persist.LoginName = resp.Login.LoginName
 	}
 
@@ -474,7 +524,7 @@ func (c *Direct) doLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags Logi
 	return false, resp.AuthURL, nil
 }
 
-func sameStrings(a, b []string) bool {
+func sameEndpoints(a, b []tailcfg.Endpoint) bool {
 	if len(a) != len(b) {
 		return false
 	}
@@ -490,15 +540,19 @@ func sameStrings(a, b []string) bool {
 // whether they've changed.
 //
 // It does not retain the provided slice.
-func (c *Direct) newEndpoints(localPort uint16, endpoints []string) (changed bool) {
+func (c *Direct) newEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) (changed bool) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
 	// Nothing new?
-	if c.localPort == localPort && sameStrings(c.endpoints, endpoints) {
+	if c.localPort == localPort && sameEndpoints(c.endpoints, endpoints) {
 		return false // unchanged
 	}
-	c.logf("client.newEndpoints(%v, %v)", localPort, endpoints)
+	var epStrs []string
+	for _, ep := range endpoints {
+		epStrs = append(epStrs, ep.Addr.String())
+	}
+	c.logf("client.newEndpoints(%v, %v)", localPort, epStrs)
 	c.localPort = localPort
 	c.endpoints = append(c.endpoints[:0], endpoints...)
 	if len(endpoints) > 0 {
@@ -510,7 +564,7 @@ func (c *Direct) newEndpoints(localPort uint16, endpoints []string) (changed boo
 // SetEndpoints updates the list of locally advertised endpoints.
 // It won't be replicated to the server until a *fresh* call to PollNetMap().
 // You don't need to restart PollNetMap if we return changed==false.
-func (c *Direct) SetEndpoints(localPort uint16, endpoints []string) (changed bool) {
+func (c *Direct) SetEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) (changed bool) {
 	// (no log message on function entry, because it clutters the logs
 	//  if endpoints haven't changed. newEndpoints() will log it.)
 	return c.newEndpoints(localPort, endpoints)
@@ -534,6 +588,11 @@ func (c *Direct) SendLiteMapUpdate(ctx context.Context) error {
 	return c.sendMapRequest(ctx, 1, nil)
 }
 
+// If we go more than pollTimeout without hearing from the server,
+// end the long poll. We should be receiving a keep alive ping
+// every minute.
+const pollTimeout = 120 * time.Second
+
 // cb nil means to omit peers.
 func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netmap.NetworkMap)) error {
 	c.mu.Lock()
@@ -543,7 +602,12 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 	hostinfo := c.hostinfo.Clone()
 	backendLogID := hostinfo.BackendLogID
 	localPort := c.localPort
-	ep := append([]string(nil), c.endpoints...)
+	var epStrs []string
+	var epTypes []tailcfg.EndpointType
+	for _, ep := range c.endpoints {
+		epStrs = append(epStrs, ep.Addr.String())
+		epTypes = append(epTypes, ep.Type)
+	}
 	everEndpoints := c.everEndpoints
 	c.mu.Unlock()
 
@@ -563,7 +627,7 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 	}
 
 	allowStream := maxPolls != 1
-	c.logf("[v1] PollNetMap: stream=%v :%v ep=%v", allowStream, localPort, ep)
+	c.logf("[v1] PollNetMap: stream=%v :%v ep=%v", allowStream, localPort, epStrs)
 
 	vlogf := logger.Discard
 	if Debug.NetMap {
@@ -573,15 +637,16 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 	}
 
 	request := &tailcfg.MapRequest{
-		Version:    tailcfg.CurrentMapRequestVersion,
-		KeepAlive:  c.keepAlive,
-		NodeKey:    tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
-		DiscoKey:   c.discoPubKey,
-		Endpoints:  ep,
-		Stream:     allowStream,
-		Hostinfo:   hostinfo,
-		DebugFlags: c.debugFlags,
-		OmitPeers:  cb == nil,
+		Version:       tailcfg.CurrentMapRequestVersion,
+		KeepAlive:     c.keepAlive,
+		NodeKey:       tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
+		DiscoKey:      c.discoPubKey,
+		Endpoints:     epStrs,
+		EndpointTypes: epTypes,
+		Stream:        allowStream,
+		Hostinfo:      hostinfo,
+		DebugFlags:    c.debugFlags,
+		OmitPeers:     cb == nil,
 	}
 	var extraDebugFlags []string
 	if hostinfo != nil && c.linkMon != nil && !c.skipIPForwardingCheck &&
@@ -609,7 +674,7 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 	// TODO(bradfitz): we skip this optimization in tests, though,
 	// because the e2e tests are currently hyperspecific about the
 	// ordering of things. The e2e tests need love.
-	if len(ep) == 0 && !everEndpoints && !inTest() {
+	if len(epStrs) == 0 && !everEndpoints && !inTest() {
 		request.ReadOnly = true
 	}
 
@@ -652,10 +717,6 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		return nil
 	}
 
-	// If we go more than pollTimeout without hearing from the server,
-	// end the long poll. We should be receiving a keep alive ping
-	// every minute.
-	const pollTimeout = 120 * time.Second
 	timeout := time.NewTimer(pollTimeout)
 	timeoutReset := make(chan struct{})
 	pollDone := make(chan struct{})
@@ -685,10 +746,11 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		}
 	}()
 
-	var lastDERPMap *tailcfg.DERPMap
-	var lastUserProfile = map[tailcfg.UserID]tailcfg.UserProfile{}
-	var lastParsedPacketFilter []filter.Match
-	var collectServices bool
+	sess := newMapSession(persist.PrivateNodeKey)
+	sess.logf = c.logf
+	sess.vlogf = vlogf
+	sess.machinePubKey = machinePubKey
+	sess.keepSharerAndUserSplit = c.keepSharerAndUserSplit
 
 	// If allowStream, then the server will use an HTTP long poll to
 	// return incremental results. There is always one response right
@@ -697,7 +759,6 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 	// the same format before just closing the connection.
 	// We can use this same read loop either way.
 	var msg []byte
-	var previousPeers []*tailcfg.Node // for delta-purposes
 	for i := 0; i < maxPolls || maxPolls < 0; i++ {
 		vlogf("netmap: starting size read after %v (poll %v)", time.Since(t0).Round(time.Millisecond), i)
 		var siz [4]byte
@@ -724,7 +785,7 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			health.GotStreamedMapResponse()
 		}
 
-		if pr := resp.PingRequest; pr != nil {
+		if pr := resp.PingRequest; pr != nil && c.isUniquePingRequest(pr) {
 			go answerPing(c.logf, c.httpc, pr)
 		}
 
@@ -744,17 +805,10 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			continue
 		}
 
-		undeltaPeers(&resp, previousPeers)
-		previousPeers = cloneNodes(resp.Peers) // defensive/lazy clone, since this escapes to who knows where
-		for _, up := range resp.UserProfiles {
-			lastUserProfile[up.ID] = up
-		}
-
-		if resp.DERPMap != nil {
-			vlogf("netmap: new map contains DERP map")
-			lastDERPMap = resp.DERPMap
-		}
-		if resp.Debug != nil {
+		hasDebug := resp.Debug != nil
+		// being conservative here, if Debug not present set to False
+		controlknobs.SetDisableUPnP(hasDebug && resp.Debug.DisableUPnP.EqualBool(true))
+		if hasDebug {
 			if resp.Debug.LogHeapPprof {
 				go logheap.LogHeap(resp.Debug.LogHeapURL)
 			}
@@ -763,18 +817,40 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			}
 			setControlAtomic(&controlUseDERPRoute, resp.Debug.DERPRoute)
 			setControlAtomic(&controlTrimWGConfig, resp.Debug.TrimWGConfig)
+			if sleep := time.Duration(resp.Debug.SleepSeconds * float64(time.Second)); sleep > 0 {
+				if err := sleepAsRequested(ctx, c.logf, timeoutReset, sleep); err != nil {
+					return err
+				}
+			}
 		}
+
+		nm := sess.netmapForResponse(&resp)
+		if nm.SelfNode == nil {
+			c.logf("MapResponse lacked node")
+			return errors.New("MapResponse lacked node")
+		}
+
 		// Temporarily (2020-06-29) support removing all but
 		// discovery-supporting nodes during development, for
 		// less noise.
 		if Debug.OnlyDisco {
-			filtered := resp.Peers[:0]
-			for _, p := range resp.Peers {
-				if !p.DiscoKey.IsZero() {
-					filtered = append(filtered, p)
+			anyOld, numDisco := false, 0
+			for _, p := range nm.Peers {
+				if p.DiscoKey.IsZero() {
+					anyOld = true
+				} else {
+					numDisco++
 				}
 			}
-			resp.Peers = filtered
+			if anyOld {
+				filtered := make([]*tailcfg.Node, 0, numDisco)
+				for _, p := range nm.Peers {
+					if !p.DiscoKey.IsZero() {
+						filtered = append(filtered, p)
+					}
+				}
+				nm.Peers = filtered
+			}
 		}
 		if Debug.StripEndpoints {
 			for _, p := range resp.Peers {
@@ -784,13 +860,8 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 				p.Endpoints = []string{"127.9.9.9:456"}
 			}
 		}
-
-		if pf := resp.PacketFilter; pf != nil {
-			lastParsedPacketFilter = c.parsePacketFilter(pf)
-		}
-
-		if v, ok := resp.CollectServices.Get(); ok {
-			collectServices = v
+		if Debug.StripCaps {
+			nm.SelfNode.Capabilities = nil
 		}
 
 		// Get latest localPort. This might've changed if
@@ -798,67 +869,9 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		// the end-to-end test.
 		// TODO(bradfitz): remove the NetworkMap.LocalPort field entirely.
 		c.mu.Lock()
-		localPort = c.localPort
+		nm.LocalPort = c.localPort
 		c.mu.Unlock()
 
-		nm := &netmap.NetworkMap{
-			SelfNode:        resp.Node,
-			NodeKey:         tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
-			PrivateKey:      persist.PrivateNodeKey,
-			MachineKey:      machinePubKey,
-			Expiry:          resp.Node.KeyExpiry,
-			Name:            resp.Node.Name,
-			Addresses:       resp.Node.Addresses,
-			Peers:           resp.Peers,
-			LocalPort:       localPort,
-			User:            resp.Node.User,
-			UserProfiles:    make(map[tailcfg.UserID]tailcfg.UserProfile),
-			Domain:          resp.Domain,
-			DNS:             resp.DNSConfig,
-			Hostinfo:        resp.Node.Hostinfo,
-			PacketFilter:    lastParsedPacketFilter,
-			CollectServices: collectServices,
-			DERPMap:         lastDERPMap,
-			Debug:           resp.Debug,
-		}
-		addUserProfile := func(userID tailcfg.UserID) {
-			if _, dup := nm.UserProfiles[userID]; dup {
-				// Already populated it from a previous peer.
-				return
-			}
-			if up, ok := lastUserProfile[userID]; ok {
-				nm.UserProfiles[userID] = up
-			}
-		}
-		addUserProfile(nm.User)
-		magicDNSSuffix := nm.MagicDNSSuffix()
-		nm.SelfNode.InitDisplayNames(magicDNSSuffix)
-		for _, peer := range resp.Peers {
-			peer.InitDisplayNames(magicDNSSuffix)
-			if !peer.Sharer.IsZero() {
-				if c.keepSharerAndUserSplit {
-					addUserProfile(peer.Sharer)
-				} else {
-					peer.User = peer.Sharer
-				}
-			}
-			addUserProfile(peer.User)
-		}
-		if resp.Node.MachineAuthorized {
-			nm.MachineStatus = tailcfg.MachineAuthorized
-		} else {
-			nm.MachineStatus = tailcfg.MachineUnauthorized
-		}
-		if len(resp.DNS) > 0 {
-			nm.DNS.Nameservers = resp.DNS
-		}
-		if len(resp.SearchPaths) > 0 {
-			nm.DNS.Domains = resp.SearchPaths
-		}
-		if Debug.ProxyDNS {
-			nm.DNS.Proxied = true
-		}
-
 		// Printing the netmap can be extremely verbose, but is very
 		// handy for debugging. Let's limit how often we do it.
 		// Code elsewhere prints netmap diffs every time, so this
@@ -894,7 +907,10 @@ func decode(res *http.Response, v interface{}, serverKey *wgkey.Key, mkey *wgkey
 	return decodeMsg(msg, v, serverKey, mkey)
 }
 
-var debugMap, _ = strconv.ParseBool(os.Getenv("TS_DEBUG_MAP"))
+var (
+	debugMap, _      = strconv.ParseBool(os.Getenv("TS_DEBUG_MAP"))
+	debugRegister, _ = strconv.ParseBool(os.Getenv("TS_DEBUG_REGISTER"))
+)
 
 var jsonEscapedZero = []byte(`\u0000`)
 
@@ -1021,6 +1037,7 @@ type debug struct {
 	OnlyDisco      bool
 	Disco          bool
 	StripEndpoints bool // strip endpoints from control (only use disco messages)
+	StripCaps      bool // strip all local node's control-provided capabilities
 }
 
 func initDebug() debug {
@@ -1029,6 +1046,7 @@ func initDebug() debug {
 		NetMap:         envBool("TS_DEBUG_NETMAP"),
 		ProxyDNS:       envBool("TS_DEBUG_PROXY_DNS"),
 		StripEndpoints: envBool("TS_DEBUG_STRIP_ENDPOINTS"),
+		StripCaps:      envBool("TS_DEBUG_STRIP_CAPS"),
 		OnlyDisco:      use == "only",
 		Disco:          use == "only" || use == "" || envBool("TS_DEBUG_USE_DISCO"),
 	}
@@ -1046,117 +1064,7 @@ func envBool(k string) bool {
 	return v
 }
 
-// undeltaPeers updates mapRes.Peers to be complete based on the provided previous peer list
-// and the PeersRemoved and PeersChanged fields in mapRes.
-// It then also nils out the delta fields.
-func undeltaPeers(mapRes *tailcfg.MapResponse, prev []*tailcfg.Node) {
-	if len(mapRes.Peers) > 0 {
-		// Not delta encoded.
-		if !nodesSorted(mapRes.Peers) {
-			log.Printf("netmap: undeltaPeers: MapResponse.Peers not sorted; sorting")
-			sortNodes(mapRes.Peers)
-		}
-		return
-	}
-
-	var removed map[tailcfg.NodeID]bool
-	if pr := mapRes.PeersRemoved; len(pr) > 0 {
-		removed = make(map[tailcfg.NodeID]bool, len(pr))
-		for _, id := range pr {
-			removed[id] = true
-		}
-	}
-	changed := mapRes.PeersChanged
-
-	if len(removed) == 0 && len(changed) == 0 {
-		// No changes fast path.
-		mapRes.Peers = prev
-		return
-	}
-
-	if !nodesSorted(changed) {
-		log.Printf("netmap: undeltaPeers: MapResponse.PeersChanged not sorted; sorting")
-		sortNodes(changed)
-	}
-	if !nodesSorted(prev) {
-		// Internal error (unrelated to the network) if we get here.
-		log.Printf("netmap: undeltaPeers: [unexpected] prev not sorted; sorting")
-		sortNodes(prev)
-	}
-
-	newFull := make([]*tailcfg.Node, 0, len(prev)-len(removed))
-	for len(prev) > 0 && len(changed) > 0 {
-		pID := prev[0].ID
-		cID := changed[0].ID
-		if removed[pID] {
-			prev = prev[1:]
-			continue
-		}
-		switch {
-		case pID < cID:
-			newFull = append(newFull, prev[0])
-			prev = prev[1:]
-		case pID == cID:
-			newFull = append(newFull, changed[0])
-			prev, changed = prev[1:], changed[1:]
-		case cID < pID:
-			newFull = append(newFull, changed[0])
-			changed = changed[1:]
-		}
-	}
-	newFull = append(newFull, changed...)
-	for _, n := range prev {
-		if !removed[n.ID] {
-			newFull = append(newFull, n)
-		}
-	}
-	sortNodes(newFull)
-
-	if mapRes.PeerSeenChange != nil {
-		peerByID := make(map[tailcfg.NodeID]*tailcfg.Node, len(newFull))
-		for _, n := range newFull {
-			peerByID[n.ID] = n
-		}
-		now := time.Now()
-		for nodeID, seen := range mapRes.PeerSeenChange {
-			if n, ok := peerByID[nodeID]; ok {
-				if seen {
-					n.LastSeen = &now
-				} else {
-					n.LastSeen = nil
-				}
-			}
-		}
-	}
-
-	mapRes.Peers = newFull
-	mapRes.PeersChanged = nil
-	mapRes.PeersRemoved = nil
-}
-
-func nodesSorted(v []*tailcfg.Node) bool {
-	for i, n := range v {
-		if i > 0 && n.ID <= v[i-1].ID {
-			return false
-		}
-	}
-	return true
-}
-
-func sortNodes(v []*tailcfg.Node) {
-	sort.Slice(v, func(i, j int) bool { return v[i].ID < v[j].ID })
-}
-
-func cloneNodes(v1 []*tailcfg.Node) []*tailcfg.Node {
-	if v1 == nil {
-		return nil
-	}
-	v2 := make([]*tailcfg.Node, len(v1))
-	for i, n := range v1 {
-		v2[i] = n.Clone()
-	}
-	return v2
-}
+var clockNow = time.Now
 
 // opt.Bool configs from control.
 var (
@@ -1211,7 +1119,7 @@ func ipForwardingBroken(routes []netaddr.IPPrefix, state *interfaces.State) bool
 	localIPs := map[netaddr.IP]bool{}
 	for _, addrs := range state.InterfaceIPs {
 		for _, pfx := range addrs {
-			localIPs[pfx.IP] = true
+			localIPs[pfx.IP()] = true
 		}
 	}
 
@@ -1220,10 +1128,10 @@ func ipForwardingBroken(routes []netaddr.IPPrefix, state *interfaces.State) bool
 		// It's possible to advertise a route to one of the local
 		// machine's local IPs. IP forwarding isn't required for this
 		// to work, so we shouldn't warn for such exports.
-		if r.IsSingleIP() && localIPs[r.IP] {
+		if r.IsSingleIP() && localIPs[r.IP()] {
 			continue
 		}
-		if r.IP.Is4() {
+		if r.IP().Is4() {
 			v4Routes = true
 		} else {
 			v6Routes = true
@@ -1275,6 +1183,23 @@ func ipForwardingBroken(routes []netaddr.IPPrefix, state *interfaces.State) bool
 	return false
 }
 
+// isUniquePingRequest reports whether pr contains a new PingRequest.URL
+// not already handled, noting its value when returning true.
+func (c *Direct) isUniquePingRequest(pr *tailcfg.PingRequest) bool {
+	if pr == nil || pr.URL == "" {
+		// Bogus.
+		return false
+	}
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if pr.URL == c.lastPingURL {
+		return false
+	}
+	c.lastPingURL = pr.URL
+	return true
+}
+
 func answerPing(logf logger.Logf, c *http.Client, pr *tailcfg.PingRequest) {
 	if pr.URL == "" {
 		logf("invalid PingRequest with no URL")
@@ -1300,3 +1225,81 @@ func answerPing(logf logger.Logf, c *http.Client, pr *tailcfg.PingRequest) {
 		logf("answerPing complete to %v (after %v)", pr.URL, d)
 	}
 }
+
+func sleepAsRequested(ctx context.Context, logf logger.Logf, timeoutReset chan<- struct{}, d time.Duration) error {
+	const maxSleep = 5 * time.Minute
+	if d > maxSleep {
+		logf("sleeping for %v, capped from server-requested %v ...", maxSleep, d)
+		d = maxSleep
+	} else {
+		logf("sleeping for server-requested %v ...", d)
+	}
+
+	ticker := time.NewTicker(pollTimeout / 2)
+	defer ticker.Stop()
+	timer := time.NewTimer(d)
+	defer timer.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case <-timer.C:
+			return nil
+		case <-ticker.C:
+			select {
+			case timeoutReset <- struct{}{}:
+			case <-timer.C:
+				return nil
+			case <-ctx.Done():
+				return ctx.Err()
+			}
+		}
+	}
+}
+
+// SetDNS sends the SetDNSRequest request to the control plane server,
+// requesting a DNS record be created or updated.
+func (c *Direct) SetDNS(ctx context.Context, req *tailcfg.SetDNSRequest) error {
+	c.mu.Lock()
+	serverKey := c.serverKey
+	c.mu.Unlock()
+
+	if serverKey.IsZero() {
+		return errors.New("zero serverKey")
+	}
+	machinePrivKey, err := c.getMachinePrivKey()
+	if err != nil {
+		return fmt.Errorf("getMachinePrivKey: %w", err)
+	}
+	if machinePrivKey.IsZero() {
+		return errors.New("getMachinePrivKey returned zero key")
+	}
+
+	bodyData, err := encode(req, &serverKey, &machinePrivKey)
+	if err != nil {
+		return err
+	}
+	body := bytes.NewReader(bodyData)
+
+	u := fmt.Sprintf("%s/machine/%s/set-dns", c.serverURL, machinePrivKey.Public().HexString())
+	hreq, err := http.NewRequestWithContext(ctx, "POST", u, body)
+	if err != nil {
+		return err
+	}
+	res, err := c.httpc.Do(hreq)
+	if err != nil {
+		return err
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		msg, _ := ioutil.ReadAll(res.Body)
+		return fmt.Errorf("set-dns response: %v, %.200s", res.Status, strings.TrimSpace(string(msg)))
+	}
+	var setDNSRes struct{} // no fields yet
+	if err := decode(res, &setDNSRes, &serverKey, &machinePrivKey); err != nil {
+		c.logf("error decoding SetDNSResponse with server key %s and machine key %s: %v", serverKey, machinePrivKey.Public(), err)
+		return fmt.Errorf("set-dns-response: %v", err)
+	}
+
+	return nil
+}

control/controlclient/direct_test.go

@@ -6,94 +6,13 @@ package controlclient
 
 import (
 	"encoding/json"
-	"fmt"
-	"reflect"
-	"strings"
 	"testing"
 
+	"inet.af/netaddr"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/wgkey"
 )
 
-func TestUndeltaPeers(t *testing.T) {
-	n := func(id tailcfg.NodeID, name string) *tailcfg.Node {
-		return &tailcfg.Node{ID: id, Name: name}
-	}
-	peers := func(nv ...*tailcfg.Node) []*tailcfg.Node { return nv }
-	tests := []struct {
-		name   string
-		mapRes *tailcfg.MapResponse
-		prev   []*tailcfg.Node
-		want   []*tailcfg.Node
-	}{
-		{
-			name: "full_peers",
-			mapRes: &tailcfg.MapResponse{
-				Peers: peers(n(1, "foo"), n(2, "bar")),
-			},
-			want: peers(n(1, "foo"), n(2, "bar")),
-		},
-		{
-			name: "full_peers_ignores_deltas",
-			mapRes: &tailcfg.MapResponse{
-				Peers:        peers(n(1, "foo"), n(2, "bar")),
-				PeersRemoved: []tailcfg.NodeID{2},
-			},
-			want: peers(n(1, "foo"), n(2, "bar")),
-		},
-		{
-			name: "add_and_update",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				PeersChanged: peers(n(0, "zero"), n(2, "bar2"), n(3, "three")),
-			},
-			want: peers(n(0, "zero"), n(1, "foo"), n(2, "bar2"), n(3, "three")),
-		},
-		{
-			name: "remove",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				PeersRemoved: []tailcfg.NodeID{1},
-			},
-			want: peers(n(2, "bar")),
-		},
-		{
-			name: "add_and_remove",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				PeersChanged: peers(n(1, "foo2")),
-				PeersRemoved: []tailcfg.NodeID{2},
-			},
-			want: peers(n(1, "foo2")),
-		},
-		{
-			name:   "unchanged",
-			prev:   peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{},
-			want:   peers(n(1, "foo"), n(2, "bar")),
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			undeltaPeers(tt.mapRes, tt.prev)
-			if !reflect.DeepEqual(tt.mapRes.Peers, tt.want) {
-				t.Errorf("wrong results\n got: %s\nwant: %s", formatNodes(tt.mapRes.Peers), formatNodes(tt.want))
-			}
-		})
-	}
-}
-
-func formatNodes(nodes []*tailcfg.Node) string {
-	var sb strings.Builder
-	for i, n := range nodes {
-		if i > 0 {
-			sb.WriteString(", ")
-		}
-		fmt.Fprintf(&sb, "(%d, %q)", n.ID, n.Name)
-	}
-	return sb.String()
-}
-
 func TestNewDirect(t *testing.T) {
 	hi := NewHostinfo()
 	ni := tailcfg.NetInfo{LinkType: "wired"}
@@ -144,7 +63,7 @@ func TestNewDirect(t *testing.T) {
 		t.Errorf("c.SetHostinfo(hi) want true got %v", changed)
 	}
 
-	endpoints := []string{"1", "2", "3"}
+	endpoints := fakeEndpoints(1, 2, 3)
 	changed = c.newEndpoints(12, endpoints)
 	if !changed {
 		t.Errorf("c.newEndpoints(12) want true got %v", changed)
@@ -157,13 +76,22 @@ func TestNewDirect(t *testing.T) {
 	if !changed {
 		t.Errorf("c.newEndpoints(13) want true got %v", changed)
 	}
-	endpoints = []string{"4", "5", "6"}
+	endpoints = fakeEndpoints(4, 5, 6)
 	changed = c.newEndpoints(13, endpoints)
 	if !changed {
 		t.Errorf("c.newEndpoints(13) want true got %v", changed)
 	}
 }
 
+func fakeEndpoints(ports ...uint16) (ret []tailcfg.Endpoint) {
+	for _, port := range ports {
+		ret = append(ret, tailcfg.Endpoint{
+			Addr: netaddr.IPPortFrom(netaddr.IP{}, port),
+		})
+	}
+	return
+}
+
 func TestNewHostinfo(t *testing.T) {
 	hi := NewHostinfo()
 	if hi == nil {

control/controlclient/filter.go

@@ -1,20 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package controlclient
-
-import (
-	"tailscale.com/tailcfg"
-	"tailscale.com/wgengine/filter"
-)
-
-// Parse a backward-compatible FilterRule used by control's wire
-// format, producing the most current filter format.
-func (c *Direct) parsePacketFilter(pf []tailcfg.FilterRule) []filter.Match {
-	mm, err := filter.MatchesFromFilterRules(pf)
-	if err != nil {
-		c.logf("parsePacketFilter: %s\n", err)
-	}
-	return mm
-}

control/controlclient/hostinfo_linux.go

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build linux && !android
 // +build linux,!android
 
 package controlclient
@@ -9,12 +10,11 @@ package controlclient
 import (
 	"bytes"
 	"fmt"
-	"io"
 	"io/ioutil"
 	"strings"
 	"syscall"
 
-	"go4.org/mem"
+	"tailscale.com/hostinfo"
 	"tailscale.com/util/lineread"
 	"tailscale.com/version/distro"
 )
@@ -55,9 +55,12 @@ func osVersionLinux() string {
 		}
 		attrBuf.WriteByte(byte(b))
 	}
-	if inContainer() {
+	if hostinfo.InContainer() {
 		attrBuf.WriteString("; container")
 	}
+	if env := hostinfo.GetEnvType(); env != "" {
+		fmt.Fprintf(&attrBuf, "; env=%s", env)
+	}
 	attr := attrBuf.String()
 
 	id := m["ID"]
@@ -89,15 +92,3 @@ func osVersionLinux() string {
 	}
 	return fmt.Sprintf("Other%s", attr)
 }
-
-func inContainer() (ret bool) {
-	lineread.File("/proc/1/cgroup", func(line []byte) error {
-		if mem.Contains(mem.B(line), mem.S("/docker/")) ||
-			mem.Contains(mem.B(line), mem.S("/lxc/")) {
-			ret = true
-			return io.EOF // arbitrary non-nil error to stop loop
-		}
-		return nil
-	})
-	return
-}

control/controlclient/map.go

@@ -0,0 +1,301 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package controlclient
+
+import (
+	"log"
+	"os"
+	"sort"
+	"strconv"
+
+	"inet.af/netaddr"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/wgkey"
+	"tailscale.com/wgengine/filter"
+)
+
+// mapSession holds the state over a long-polled "map" request to the
+// control plane.
+//
+// It accepts incremental tailcfg.MapResponse values to
+// netMapForResponse and returns fully inflated NetworkMaps, filling
+// in the omitted data implicit from prior MapResponse values from
+// within the same session (the same long-poll HTTP response to the
+// one MapRequest).
+type mapSession struct {
+	// Immutable fields.
+	privateNodeKey         wgkey.Private
+	logf                   logger.Logf
+	vlogf                  logger.Logf
+	machinePubKey          tailcfg.MachineKey
+	keepSharerAndUserSplit bool // see Options.KeepSharerAndUserSplit
+
+	// Fields storing state over the the coards of multiple MapResponses.
+	lastNode               *tailcfg.Node
+	lastDNSConfig          *tailcfg.DNSConfig
+	lastDERPMap            *tailcfg.DERPMap
+	lastUserProfile        map[tailcfg.UserID]tailcfg.UserProfile
+	lastParsedPacketFilter []filter.Match
+	collectServices        bool
+	previousPeers          []*tailcfg.Node // for delta-purposes
+	lastDomain             string
+
+	// netMapBuilding is non-nil during a netmapForResponse call,
+	// containing the value to be returned, once fully populated.
+	netMapBuilding *netmap.NetworkMap
+}
+
+func newMapSession(privateNodeKey wgkey.Private) *mapSession {
+	ms := &mapSession{
+		privateNodeKey:  privateNodeKey,
+		logf:            logger.Discard,
+		vlogf:           logger.Discard,
+		lastDNSConfig:   new(tailcfg.DNSConfig),
+		lastUserProfile: map[tailcfg.UserID]tailcfg.UserProfile{},
+	}
+	return ms
+}
+
+func (ms *mapSession) addUserProfile(userID tailcfg.UserID) {
+	nm := ms.netMapBuilding
+	if _, dup := nm.UserProfiles[userID]; dup {
+		// Already populated it from a previous peer.
+		return
+	}
+	if up, ok := ms.lastUserProfile[userID]; ok {
+		nm.UserProfiles[userID] = up
+	}
+}
+
+// netmapForResponse returns a fully populated NetworkMap from a full
+// or incremental MapResponse within the session, filling in omitted
+// information from prior MapResponse values.
+func (ms *mapSession) netmapForResponse(resp *tailcfg.MapResponse) *netmap.NetworkMap {
+	undeltaPeers(resp, ms.previousPeers)
+
+	ms.previousPeers = cloneNodes(resp.Peers) // defensive/lazy clone, since this escapes to who knows where
+	for _, up := range resp.UserProfiles {
+		ms.lastUserProfile[up.ID] = up
+	}
+
+	if resp.DERPMap != nil {
+		ms.vlogf("netmap: new map contains DERP map")
+		ms.lastDERPMap = resp.DERPMap
+	}
+
+	if pf := resp.PacketFilter; pf != nil {
+		var err error
+		ms.lastParsedPacketFilter, err = filter.MatchesFromFilterRules(pf)
+		if err != nil {
+			ms.logf("parsePacketFilter: %v", err)
+		}
+	}
+	if c := resp.DNSConfig; c != nil {
+		ms.lastDNSConfig = c
+	}
+
+	if v, ok := resp.CollectServices.Get(); ok {
+		ms.collectServices = v
+	}
+	if resp.Domain != "" {
+		ms.lastDomain = resp.Domain
+	}
+
+	nm := &netmap.NetworkMap{
+		NodeKey:         tailcfg.NodeKey(ms.privateNodeKey.Public()),
+		PrivateKey:      ms.privateNodeKey,
+		MachineKey:      ms.machinePubKey,
+		Peers:           resp.Peers,
+		UserProfiles:    make(map[tailcfg.UserID]tailcfg.UserProfile),
+		Domain:          ms.lastDomain,
+		DNS:             *ms.lastDNSConfig,
+		PacketFilter:    ms.lastParsedPacketFilter,
+		CollectServices: ms.collectServices,
+		DERPMap:         ms.lastDERPMap,
+		Debug:           resp.Debug,
+	}
+	ms.netMapBuilding = nm
+
+	if resp.Node != nil {
+		ms.lastNode = resp.Node
+	}
+	if node := ms.lastNode.Clone(); node != nil {
+		nm.SelfNode = node
+		nm.Expiry = node.KeyExpiry
+		nm.Name = node.Name
+		nm.Addresses = filterSelfAddresses(node.Addresses)
+		nm.User = node.User
+		nm.Hostinfo = node.Hostinfo
+		if node.MachineAuthorized {
+			nm.MachineStatus = tailcfg.MachineAuthorized
+		} else {
+			nm.MachineStatus = tailcfg.MachineUnauthorized
+		}
+	}
+
+	ms.addUserProfile(nm.User)
+	magicDNSSuffix := nm.MagicDNSSuffix()
+	if nm.SelfNode != nil {
+		nm.SelfNode.InitDisplayNames(magicDNSSuffix)
+	}
+	for _, peer := range resp.Peers {
+		peer.InitDisplayNames(magicDNSSuffix)
+		if !peer.Sharer.IsZero() {
+			if ms.keepSharerAndUserSplit {
+				ms.addUserProfile(peer.Sharer)
+			} else {
+				peer.User = peer.Sharer
+			}
+		}
+		ms.addUserProfile(peer.User)
+	}
+	if len(resp.DNS) > 0 {
+		nm.DNS.Nameservers = resp.DNS
+	}
+	if len(resp.SearchPaths) > 0 {
+		nm.DNS.Domains = resp.SearchPaths
+	}
+	if Debug.ProxyDNS {
+		nm.DNS.Proxied = true
+	}
+	ms.netMapBuilding = nil
+	return nm
+}
+
+// undeltaPeers updates mapRes.Peers to be complete based on the
+// provided previous peer list and the PeersRemoved and PeersChanged
+// fields in mapRes, as well as the PeerSeenChange and OnlineChange
+// maps.
+//
+// It then also nils out the delta fields.
+func undeltaPeers(mapRes *tailcfg.MapResponse, prev []*tailcfg.Node) {
+	if len(mapRes.Peers) > 0 {
+		// Not delta encoded.
+		if !nodesSorted(mapRes.Peers) {
+			log.Printf("netmap: undeltaPeers: MapResponse.Peers not sorted; sorting")
+			sortNodes(mapRes.Peers)
+		}
+		return
+	}
+
+	var removed map[tailcfg.NodeID]bool
+	if pr := mapRes.PeersRemoved; len(pr) > 0 {
+		removed = make(map[tailcfg.NodeID]bool, len(pr))
+		for _, id := range pr {
+			removed[id] = true
+		}
+	}
+	changed := mapRes.PeersChanged
+
+	if !nodesSorted(changed) {
+		log.Printf("netmap: undeltaPeers: MapResponse.PeersChanged not sorted; sorting")
+		sortNodes(changed)
+	}
+	if !nodesSorted(prev) {
+		// Internal error (unrelated to the network) if we get here.
+		log.Printf("netmap: undeltaPeers: [unexpected] prev not sorted; sorting")
+		sortNodes(prev)
+	}
+
+	newFull := prev
+	if len(removed) > 0 || len(changed) > 0 {
+		newFull = make([]*tailcfg.Node, 0, len(prev)-len(removed))
+		for len(prev) > 0 && len(changed) > 0 {
+			pID := prev[0].ID
+			cID := changed[0].ID
+			if removed[pID] {
+				prev = prev[1:]
+				continue
+			}
+			switch {
+			case pID < cID:
+				newFull = append(newFull, prev[0])
+				prev = prev[1:]
+			case pID == cID:
+				newFull = append(newFull, changed[0])
+				prev, changed = prev[1:], changed[1:]
+			case cID < pID:
+				newFull = append(newFull, changed[0])
+				changed = changed[1:]
+			}
+		}
+		newFull = append(newFull, changed...)
+		for _, n := range prev {
+			if !removed[n.ID] {
+				newFull = append(newFull, n)
+			}
+		}
+		sortNodes(newFull)
+	}
+
+	if len(mapRes.PeerSeenChange) != 0 || len(mapRes.OnlineChange) != 0 {
+		peerByID := make(map[tailcfg.NodeID]*tailcfg.Node, len(newFull))
+		for _, n := range newFull {
+			peerByID[n.ID] = n
+		}
+		now := clockNow()
+		for nodeID, seen := range mapRes.PeerSeenChange {
+			if n, ok := peerByID[nodeID]; ok {
+				if seen {
+					n.LastSeen = &now
+				} else {
+					n.LastSeen = nil
+				}
+			}
+		}
+		for nodeID, online := range mapRes.OnlineChange {
+			if n, ok := peerByID[nodeID]; ok {
+				online := online
+				n.Online = &online
+			}
+		}
+	}
+
+	mapRes.Peers = newFull
+	mapRes.PeersChanged = nil
+	mapRes.PeersRemoved = nil
+}
+
+func nodesSorted(v []*tailcfg.Node) bool {
+	for i, n := range v {
+		if i > 0 && n.ID <= v[i-1].ID {
+			return false
+		}
+	}
+	return true
+}
+
+func sortNodes(v []*tailcfg.Node) {
+	sort.Slice(v, func(i, j int) bool { return v[i].ID < v[j].ID })
+}
+
+func cloneNodes(v1 []*tailcfg.Node) []*tailcfg.Node {
+	if v1 == nil {
+		return nil
+	}
+	v2 := make([]*tailcfg.Node, len(v1))
+	for i, n := range v1 {
+		v2[i] = n.Clone()
+	}
+	return v2
+}
+
+var debugSelfIPv6Only, _ = strconv.ParseBool(os.Getenv("TS_DEBUG_SELF_V6_ONLY"))
+
+func filterSelfAddresses(in []netaddr.IPPrefix) (ret []netaddr.IPPrefix) {
+	switch {
+	default:
+		return in
+	case debugSelfIPv6Only:
+		for _, a := range in {
+			if a.IP().Is6() {
+				ret = append(ret, a)
+			}
+		}
+		return ret
+	}
+}

control/controlclient/map_test.go

@@ -0,0 +1,311 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package controlclient
+
+import (
+	"encoding/json"
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/wgkey"
+)
+
+func TestUndeltaPeers(t *testing.T) {
+	defer func(old func() time.Time) { clockNow = old }(clockNow)
+
+	var curTime time.Time
+	clockNow = func() time.Time {
+		return curTime
+	}
+	online := func(v bool) func(*tailcfg.Node) {
+		return func(n *tailcfg.Node) {
+			n.Online = &v
+		}
+	}
+	seenAt := func(t time.Time) func(*tailcfg.Node) {
+		return func(n *tailcfg.Node) {
+			n.LastSeen = &t
+		}
+	}
+	n := func(id tailcfg.NodeID, name string, mod ...func(*tailcfg.Node)) *tailcfg.Node {
+		n := &tailcfg.Node{ID: id, Name: name}
+		for _, f := range mod {
+			f(n)
+		}
+		return n
+	}
+	peers := func(nv ...*tailcfg.Node) []*tailcfg.Node { return nv }
+	tests := []struct {
+		name    string
+		mapRes  *tailcfg.MapResponse
+		curTime time.Time
+		prev    []*tailcfg.Node
+		want    []*tailcfg.Node
+	}{
+		{
+			name: "full_peers",
+			mapRes: &tailcfg.MapResponse{
+				Peers: peers(n(1, "foo"), n(2, "bar")),
+			},
+			want: peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "full_peers_ignores_deltas",
+			mapRes: &tailcfg.MapResponse{
+				Peers:        peers(n(1, "foo"), n(2, "bar")),
+				PeersRemoved: []tailcfg.NodeID{2},
+			},
+			want: peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "add_and_update",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersChanged: peers(n(0, "zero"), n(2, "bar2"), n(3, "three")),
+			},
+			want: peers(n(0, "zero"), n(1, "foo"), n(2, "bar2"), n(3, "three")),
+		},
+		{
+			name: "remove",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersRemoved: []tailcfg.NodeID{1},
+			},
+			want: peers(n(2, "bar")),
+		},
+		{
+			name: "add_and_remove",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersChanged: peers(n(1, "foo2")),
+				PeersRemoved: []tailcfg.NodeID{2},
+			},
+			want: peers(n(1, "foo2")),
+		},
+		{
+			name:   "unchanged",
+			prev:   peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{},
+			want:   peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "online_change",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				OnlineChange: map[tailcfg.NodeID]bool{
+					1: true,
+				},
+			},
+			want: peers(
+				n(1, "foo", online(true)),
+				n(2, "bar"),
+			),
+		},
+		{
+			name: "online_change_offline",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				OnlineChange: map[tailcfg.NodeID]bool{
+					1: false,
+					2: true,
+				},
+			},
+			want: peers(
+				n(1, "foo", online(false)),
+				n(2, "bar", online(true)),
+			),
+		},
+		{
+			name:    "peer_seen_at",
+			prev:    peers(n(1, "foo", seenAt(time.Unix(111, 0))), n(2, "bar")),
+			curTime: time.Unix(123, 0),
+			mapRes: &tailcfg.MapResponse{
+				PeerSeenChange: map[tailcfg.NodeID]bool{
+					1: false,
+					2: true,
+				},
+			},
+			want: peers(
+				n(1, "foo"),
+				n(2, "bar", seenAt(time.Unix(123, 0))),
+			),
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if !tt.curTime.IsZero() {
+				curTime = tt.curTime
+			}
+			undeltaPeers(tt.mapRes, tt.prev)
+			if !reflect.DeepEqual(tt.mapRes.Peers, tt.want) {
+				t.Errorf("wrong results\n got: %s\nwant: %s", formatNodes(tt.mapRes.Peers), formatNodes(tt.want))
+			}
+		})
+	}
+}
+
+func formatNodes(nodes []*tailcfg.Node) string {
+	var sb strings.Builder
+	for i, n := range nodes {
+		if i > 0 {
+			sb.WriteString(", ")
+		}
+		var extra string
+		if n.Online != nil {
+			extra += fmt.Sprintf(", online=%v", *n.Online)
+		}
+		if n.LastSeen != nil {
+			extra += fmt.Sprintf(", lastSeen=%v", n.LastSeen.Unix())
+		}
+		fmt.Fprintf(&sb, "(%d, %q%s)", n.ID, n.Name, extra)
+	}
+	return sb.String()
+}
+
+func newTestMapSession(t *testing.T) *mapSession {
+	k, err := wgkey.NewPrivate()
+	if err != nil {
+		t.Fatal(err)
+	}
+	return newMapSession(k)
+}
+
+func TestNetmapForResponse(t *testing.T) {
+	t.Run("implicit_packetfilter", func(t *testing.T) {
+		somePacketFilter := []tailcfg.FilterRule{
+			{
+				SrcIPs: []string{"*"},
+				DstPorts: []tailcfg.NetPortRange{
+					{IP: "10.2.3.4", Ports: tailcfg.PortRange{First: 22, Last: 22}},
+				},
+			},
+		}
+		ms := newTestMapSession(t)
+		nm1 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:         new(tailcfg.Node),
+			PacketFilter: somePacketFilter,
+		})
+		if len(nm1.PacketFilter) == 0 {
+			t.Fatalf("zero length PacketFilter")
+		}
+		nm2 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:         new(tailcfg.Node),
+			PacketFilter: nil, // testing that the server can omit this.
+		})
+		if len(nm1.PacketFilter) == 0 {
+			t.Fatalf("zero length PacketFilter in 2nd netmap")
+		}
+		if !reflect.DeepEqual(nm1.PacketFilter, nm2.PacketFilter) {
+			t.Error("packet filters differ")
+		}
+	})
+	t.Run("implicit_dnsconfig", func(t *testing.T) {
+		someDNSConfig := &tailcfg.DNSConfig{Domains: []string{"foo", "bar"}}
+		ms := newTestMapSession(t)
+		nm1 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:      new(tailcfg.Node),
+			DNSConfig: someDNSConfig,
+		})
+		if !reflect.DeepEqual(nm1.DNS, *someDNSConfig) {
+			t.Fatalf("1st DNS wrong")
+		}
+		nm2 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:      new(tailcfg.Node),
+			DNSConfig: nil, // implict
+		})
+		if !reflect.DeepEqual(nm2.DNS, *someDNSConfig) {
+			t.Fatalf("2nd DNS wrong")
+		}
+	})
+	t.Run("collect_services", func(t *testing.T) {
+		ms := newTestMapSession(t)
+		var nm *netmap.NetworkMap
+		wantCollect := func(v bool) {
+			t.Helper()
+			if nm.CollectServices != v {
+				t.Errorf("netmap.CollectServices = %v; want %v", nm.CollectServices, v)
+			}
+		}
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node: new(tailcfg.Node),
+		})
+		wantCollect(false)
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:            new(tailcfg.Node),
+			CollectServices: "false",
+		})
+		wantCollect(false)
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:            new(tailcfg.Node),
+			CollectServices: "true",
+		})
+		wantCollect(true)
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:            new(tailcfg.Node),
+			CollectServices: "",
+		})
+		wantCollect(true)
+	})
+	t.Run("implicit_domain", func(t *testing.T) {
+		ms := newTestMapSession(t)
+		var nm *netmap.NetworkMap
+		want := func(v string) {
+			t.Helper()
+			if nm.Domain != v {
+				t.Errorf("netmap.Domain = %q; want %q", nm.Domain, v)
+			}
+		}
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:   new(tailcfg.Node),
+			Domain: "foo.com",
+		})
+		want("foo.com")
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node: new(tailcfg.Node),
+		})
+		want("foo.com")
+	})
+	t.Run("implicit_node", func(t *testing.T) {
+		someNode := &tailcfg.Node{
+			Name: "foo",
+		}
+		wantNode := &tailcfg.Node{
+			Name:                 "foo",
+			ComputedName:         "foo",
+			ComputedNameWithHost: "foo",
+		}
+		ms := newTestMapSession(t)
+
+		nm1 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node: someNode,
+		})
+		if nm1.SelfNode == nil {
+			t.Fatal("nil Node in 1st netmap")
+		}
+		if !reflect.DeepEqual(nm1.SelfNode, wantNode) {
+			j, _ := json.Marshal(nm1.SelfNode)
+			t.Errorf("Node mismatch in 1st netmap; got: %s", j)
+		}
+
+		nm2 := ms.netmapForResponse(&tailcfg.MapResponse{})
+		if nm2.SelfNode == nil {
+			t.Fatal("nil Node in 1st netmap")
+		}
+		if !reflect.DeepEqual(nm2.SelfNode, wantNode) {
+			j, _ := json.Marshal(nm2.SelfNode)
+			t.Errorf("Node mismatch in 2nd netmap; got: %s", j)
+		}
+	})
+}

control/controlclient/sign_supported.go

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build windows && cgo
 // +build windows,cgo
 
 // darwin,cgo is also supported by certstore but machineCertificateSubject will
@@ -18,7 +19,7 @@ import (
 	"fmt"
 	"sync"
 
-	"github.com/github/certstore"
+	"github.com/tailscale/certstore"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/wgkey"
 	"tailscale.com/util/winutil"
@@ -51,6 +52,46 @@ var (
 	errBadRequest = errors.New("malformed request")
 )
 
+func isSupportedCertificate(cert *x509.Certificate) bool {
+	return cert.PublicKeyAlgorithm == x509.RSA
+}
+
+func isSubjectInChain(subject string, chain []*x509.Certificate) bool {
+	if len(chain) == 0 || chain[0] == nil {
+		return false
+	}
+
+	for _, c := range chain {
+		if c == nil {
+			continue
+		}
+		if c.Subject.String() == subject {
+			return true
+		}
+	}
+
+	return false
+}
+
+func selectIdentityFromSlice(subject string, ids []certstore.Identity) (certstore.Identity, []*x509.Certificate) {
+	for _, id := range ids {
+		chain, err := id.CertificateChain()
+		if err != nil {
+			continue
+		}
+
+		if !isSupportedCertificate(chain[0]) {
+			continue
+		}
+
+		if isSubjectInChain(subject, chain) {
+			return id, chain
+		}
+	}
+
+	return nil, nil
+}
+
 // findIdentity locates an identity from the Windows or Darwin certificate
 // store. It returns the first certificate with a matching Subject anywhere in
 // its certificate chain, so it is possible to search for the leaf certificate,
@@ -64,26 +105,7 @@ func findIdentity(subject string, st certstore.Store) (certstore.Identity, []*x5
 		return nil, nil, err
 	}
 
-	var selected certstore.Identity
-	var chain []*x509.Certificate
-
-	for _, id := range ids {
-		chain, err = id.CertificateChain()
-		if err != nil {
-			continue
-		}
-
-		if chain[0].PublicKeyAlgorithm != x509.RSA {
-			continue
-		}
-
-		for _, c := range chain {
-			if c.Subject.String() == subject {
-				selected = id
-				break
-			}
-		}
-	}
+	selected, chain := selectIdentityFromSlice(subject, ids)
 
 	for _, id := range ids {
 		if id != selected {

control/controlclient/sign_unsupported.go

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build !windows || !cgo
 // +build !windows !cgo
 
 package controlclient

control/controlclient/status.go

@@ -0,0 +1,105 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package controlclient
+
+import (
+	"encoding/json"
+	"fmt"
+	"reflect"
+
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/empty"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/persist"
+	"tailscale.com/types/structs"
+)
+
+// State is the high-level state of the client. It is used only in
+// unit tests for proper sequencing, don't depend on it anywhere else.
+//
+// TODO(apenwarr): eliminate the state, as it's now obsolete.
+//
+// apenwarr: Historical note: controlclient.Auto was originally
+// intended to be the state machine for the whole tailscale client, but that
+// turned out to not be the right abstraction layer, and it moved to
+// ipn.Backend. Since ipn.Backend now has a state machine, it would be
+// much better if controlclient could be a simple stateless API. But the
+// current server-side API (two interlocking polling https calls) makes that
+// very hard to implement. A server side API change could untangle this and
+// remove all the statefulness.
+type State int
+
+const (
+	StateNew = State(iota)
+	StateNotAuthenticated
+	StateAuthenticating
+	StateURLVisitRequired
+	StateAuthenticated
+	StateSynchronized // connected and received map update
+)
+
+func (s State) MarshalText() ([]byte, error) {
+	return []byte(s.String()), nil
+}
+
+func (s State) String() string {
+	switch s {
+	case StateNew:
+		return "state:new"
+	case StateNotAuthenticated:
+		return "state:not-authenticated"
+	case StateAuthenticating:
+		return "state:authenticating"
+	case StateURLVisitRequired:
+		return "state:url-visit-required"
+	case StateAuthenticated:
+		return "state:authenticated"
+	case StateSynchronized:
+		return "state:synchronized"
+	default:
+		return fmt.Sprintf("state:unknown:%d", int(s))
+	}
+}
+
+type Status struct {
+	_              structs.Incomparable
+	LoginFinished  *empty.Message // nonempty when login finishes
+	LogoutFinished *empty.Message // nonempty when logout finishes
+	Err            string
+	URL            string             // interactive URL to visit to finish logging in
+	NetMap         *netmap.NetworkMap // server-pushed configuration
+
+	// The internal state should not be exposed outside this
+	// package, but we have some automated tests elsewhere that need to
+	// use them. Please don't use these fields.
+	// TODO(apenwarr): Unexport or remove these.
+	State    State
+	Persist  *persist.Persist  // locally persisted configuration
+	Hostinfo *tailcfg.Hostinfo // current Hostinfo data
+}
+
+// Equal reports whether s and s2 are equal.
+func (s *Status) Equal(s2 *Status) bool {
+	if s == nil && s2 == nil {
+		return true
+	}
+	return s != nil && s2 != nil &&
+		(s.LoginFinished == nil) == (s2.LoginFinished == nil) &&
+		(s.LogoutFinished == nil) == (s2.LogoutFinished == nil) &&
+		s.Err == s2.Err &&
+		s.URL == s2.URL &&
+		reflect.DeepEqual(s.Persist, s2.Persist) &&
+		reflect.DeepEqual(s.NetMap, s2.NetMap) &&
+		reflect.DeepEqual(s.Hostinfo, s2.Hostinfo) &&
+		s.State == s2.State
+}
+
+func (s Status) String() string {
+	b, err := json.MarshalIndent(s, "", "\t")
+	if err != nil {
+		panic(err)
+	}
+	return s.State.String() + " " + string(b)
+}

control/controlknobs/controlknobs.go

@@ -0,0 +1,34 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package controlknobs contains client options configurable from control which can be turned on
+// or off. The ability to turn options on and off is for incrementally adding features in.
+package controlknobs
+
+import (
+	"os"
+	"strconv"
+
+	"tailscale.com/syncs"
+)
+
+// disableUPnP indicates whether to attempt UPnP mapping.
+var disableUPnP syncs.AtomicBool
+
+func init() {
+	v, _ := strconv.ParseBool(os.Getenv("TS_DISABLE_UPNP"))
+	SetDisableUPnP(v)
+}
+
+// DisableUPnP reports the last reported value from control
+// whether UPnP portmapping should be disabled.
+func DisableUPnP() bool {
+	return disableUPnP.Get()
+}
+
+// SetDisableUPnP sets whether control says that UPnP should be
+// disabled.
+func SetDisableUPnP(v bool) {
+	disableUPnP.Set(v)
+}

derp/derp_client.go

@@ -29,6 +29,7 @@ type Client struct {
 	br          *bufio.Reader
 	meshKey     string
 	canAckPings bool
+	isProber    bool
 
 	wmu sync.Mutex // hold while writing to bw
 	bw  *bufio.Writer
@@ -52,6 +53,7 @@ type clientOpt struct {
 	MeshKey     string
 	ServerPub   key.Public
 	CanAckPings bool
+	IsProber    bool
 }
 
 // MeshKey returns a ClientOpt to pass to the DERP server during connect to get
@@ -60,6 +62,10 @@ type clientOpt struct {
 // An empty key means to not use a mesh key.
 func MeshKey(key string) ClientOpt { return clientOptFunc(func(o *clientOpt) { o.MeshKey = key }) }
 
+// IsProber returns a ClientOpt to pass to the DERP server during connect to
+// declare that this client is a a prober.
+func IsProber(v bool) ClientOpt { return clientOptFunc(func(o *clientOpt) { o.IsProber = v }) }
+
 // ServerPublicKey returns a ClientOpt to declare that the server's DERP public key is known.
 // If key is the zero value, the returned ClientOpt is a no-op.
 func ServerPublicKey(key key.Public) ClientOpt {
@@ -93,6 +99,7 @@ func newClient(privateKey key.Private, nc Conn, brw *bufio.ReadWriter, logf logg
 		bw:          brw.Writer,
 		meshKey:     opt.MeshKey,
 		canAckPings: opt.CanAckPings,
+		isProber:    opt.IsProber,
 	}
 	if opt.ServerPub.IsZero() {
 		if err := c.recvServerKey(); err != nil {
@@ -160,6 +167,9 @@ type clientInfo struct {
 	// CanAckPings is whether the client declares it's able to ack
 	// pings.
 	CanAckPings bool
+
+	// IsProber is whether this client is a prober.
+	IsProber bool `json:",omitempty"`
 }
 
 func (c *Client) sendClientKey() error {
@@ -171,6 +181,7 @@ func (c *Client) sendClientKey() error {
 		Version:     ProtocolVersion,
 		MeshKey:     c.meshKey,
 		CanAckPings: c.canAckPings,
+		IsProber:    c.isProber,
 	})
 	if err != nil {
 		return err

derp/derp_server.go

@@ -20,22 +20,30 @@ import (
 	"io"
 	"io/ioutil"
 	"log"
+	"math"
 	"math/big"
 	"math/rand"
+	"net/http"
 	"os"
+	"os/exec"
 	"runtime"
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"go4.org/mem"
 	"golang.org/x/crypto/nacl/box"
 	"golang.org/x/sync/errgroup"
+	"golang.org/x/time/rate"
+	"inet.af/netaddr"
+	"tailscale.com/client/tailscale"
 	"tailscale.com/disco"
 	"tailscale.com/metrics"
 	"tailscale.com/types/key"
 	"tailscale.com/types/logger"
+	"tailscale.com/types/pad32"
 	"tailscale.com/version"
 )
 
@@ -69,13 +77,6 @@ const (
 	writeTimeout            = 2 * time.Second
 )
 
-const host64bit = (^uint(0) >> 32) & 1 // 1 on 64-bit, 0 on 32-bit
-
-// pad32bit is 4 on 32-bit machines and 0 on 64-bit.
-// It exists so the Server struct's atomic fields can be aligned to 8
-// byte boundaries. (As tested by GOARCH=386 go test, etc)
-const pad32bit = 4 - host64bit*4 // 0 on 64-bit, 4 on 32-bit
-
 // Server is a DERP server.
 type Server struct {
 	// WriteTimeout, if non-zero, specifies how long to wait
@@ -91,42 +92,46 @@ type Server struct {
 	metaCert    []byte // the encoded x509 cert to send after LetsEncrypt cert+intermediate
 
 	// Counters:
-	_                        [pad32bit]byte
-	packetsSent, bytesSent   expvar.Int
-	packetsRecv, bytesRecv   expvar.Int
-	packetsRecvByKind        metrics.LabelMap
-	packetsRecvDisco         *expvar.Int
-	packetsRecvOther         *expvar.Int
-	_                        [pad32bit]byte
-	packetsDropped           expvar.Int
-	packetsDroppedReason     metrics.LabelMap
-	packetsDroppedUnknown    *expvar.Int // unknown dst pubkey
-	packetsDroppedFwdUnknown *expvar.Int // unknown dst pubkey on forward
-	packetsDroppedGone       *expvar.Int // dst conn shutting down
-	packetsDroppedQueueHead  *expvar.Int // queue full, drop head packet
-	packetsDroppedQueueTail  *expvar.Int // queue full, drop tail packet
-	packetsDroppedWrite      *expvar.Int // error writing to dst conn
-	_                        [pad32bit]byte
-	packetsForwardedOut      expvar.Int
-	packetsForwardedIn       expvar.Int
-	peerGoneFrames           expvar.Int // number of peer gone frames sent
-	accepts                  expvar.Int
-	curClients               expvar.Int
-	curHomeClients           expvar.Int // ones with preferred
-	clientsReplaced          expvar.Int
-	unknownFrames            expvar.Int
-	homeMovesIn              expvar.Int // established clients announce home server moves in
-	homeMovesOut             expvar.Int // established clients announce home server moves out
-	multiForwarderCreated    expvar.Int
-	multiForwarderDeleted    expvar.Int
-	removePktForwardOther    expvar.Int
+	_                            pad32.Four
+	packetsSent, bytesSent       expvar.Int
+	packetsRecv, bytesRecv       expvar.Int
+	packetsRecvByKind            metrics.LabelMap
+	packetsRecvDisco             *expvar.Int
+	packetsRecvOther             *expvar.Int
+	_                            pad32.Four
+	packetsDropped               expvar.Int
+	packetsDroppedReason         metrics.LabelMap
+	packetsDroppedReasonCounters []*expvar.Int // indexed by dropReason
+	packetsDroppedType           metrics.LabelMap
+	packetsDroppedTypeDisco      *expvar.Int
+	packetsDroppedTypeOther      *expvar.Int
+	_                            pad32.Four
+	packetsForwardedOut          expvar.Int
+	packetsForwardedIn           expvar.Int
+	peerGoneFrames               expvar.Int // number of peer gone frames sent
+	accepts                      expvar.Int
+	curClients                   expvar.Int
+	curHomeClients               expvar.Int // ones with preferred
+	clientsReplaced              expvar.Int
+	clientsReplaceLimited        expvar.Int
+	clientsReplaceSleeping       expvar.Int
+	unknownFrames                expvar.Int
+	homeMovesIn                  expvar.Int // established clients announce home server moves in
+	homeMovesOut                 expvar.Int // established clients announce home server moves out
+	multiForwarderCreated        expvar.Int
+	multiForwarderDeleted        expvar.Int
+	removePktForwardOther        expvar.Int
+	avgQueueDuration             *uint64 // In milliseconds; accessed atomically
 
-	mu          sync.Mutex
-	closed      bool
-	netConns    map[Conn]chan struct{} // chan is closed when conn closes
-	clients     map[key.Public]*sclient
-	clientsEver map[key.Public]bool // never deleted from, for stats; fine for now
-	watchers    map[*sclient]bool   // mesh peer -> true
+	// verifyClients only accepts client connections to the DERP server if the clientKey is a
+	// known peer in the network, as specified by a running tailscaled's client's local api.
+	verifyClients bool
+
+	mu       sync.Mutex
+	closed   bool
+	netConns map[Conn]chan struct{} // chan is closed when conn closes
+	clients  map[key.Public]*sclient
+	watchers map[*sclient]bool // mesh peer -> true
 	// clientsMesh tracks all clients in the cluster, both locally
 	// and to mesh peers.  If the value is nil, that means the
 	// peer is only local (and thus in the clients Map, but not
@@ -138,6 +143,9 @@ type Server struct {
 	// because it includes intra-region forwarded packets as the
 	// src.
 	sentTo map[key.Public]map[key.Public]int64 // src => dst => dst's latest sclient.connNum
+
+	// maps from netaddr.IPPort to a client's public key
+	keyOfAddr map[netaddr.IPPort]key.Public
 }
 
 // PacketForwarder is something that can forward packets.
@@ -153,7 +161,7 @@ type PacketForwarder interface {
 // Conn is the subset of the underlying net.Conn the DERP Server needs.
 // It is a defined type so that non-net connections can be used.
 type Conn interface {
-	io.Closer
+	io.WriteCloser
 
 	// The *Deadline methods follow the semantics of net.Conn.
 
@@ -175,23 +183,29 @@ func NewServer(privateKey key.Private, logf logger.Logf) *Server {
 		limitedLogf:          logger.RateLimitedFn(logf, 30*time.Second, 5, 100),
 		packetsRecvByKind:    metrics.LabelMap{Label: "kind"},
 		packetsDroppedReason: metrics.LabelMap{Label: "reason"},
+		packetsDroppedType:   metrics.LabelMap{Label: "type"},
 		clients:              map[key.Public]*sclient{},
-		clientsEver:          map[key.Public]bool{},
 		clientsMesh:          map[key.Public]PacketForwarder{},
 		netConns:             map[Conn]chan struct{}{},
 		memSys0:              ms.Sys,
 		watchers:             map[*sclient]bool{},
 		sentTo:               map[key.Public]map[key.Public]int64{},
+		avgQueueDuration:     new(uint64),
+		keyOfAddr:            map[netaddr.IPPort]key.Public{},
 	}
 	s.initMetacert()
 	s.packetsRecvDisco = s.packetsRecvByKind.Get("disco")
 	s.packetsRecvOther = s.packetsRecvByKind.Get("other")
-	s.packetsDroppedUnknown = s.packetsDroppedReason.Get("unknown_dest")
-	s.packetsDroppedFwdUnknown = s.packetsDroppedReason.Get("unknown_dest_on_fwd")
-	s.packetsDroppedGone = s.packetsDroppedReason.Get("gone")
-	s.packetsDroppedQueueHead = s.packetsDroppedReason.Get("queue_head")
-	s.packetsDroppedQueueTail = s.packetsDroppedReason.Get("queue_tail")
-	s.packetsDroppedWrite = s.packetsDroppedReason.Get("write_error")
+	s.packetsDroppedReasonCounters = []*expvar.Int{
+		s.packetsDroppedReason.Get("unknown_dest"),
+		s.packetsDroppedReason.Get("unknown_dest_on_fwd"),
+		s.packetsDroppedReason.Get("gone"),
+		s.packetsDroppedReason.Get("queue_head"),
+		s.packetsDroppedReason.Get("queue_tail"),
+		s.packetsDroppedReason.Get("write_error"),
+	}
+	s.packetsDroppedTypeDisco = s.packetsDroppedType.Get("disco")
+	s.packetsDroppedTypeOther = s.packetsDroppedType.Get("other")
 	return s
 }
 
@@ -203,6 +217,13 @@ func (s *Server) SetMeshKey(v string) {
 	s.meshKey = v
 }
 
+// SetVerifyClients sets whether this DERP server verifies clients through tailscaled.
+//
+// It must be called before serving begins.
+func (s *Server) SetVerifyClient(v bool) {
+	s.verifyClients = v
+}
+
 // HasMeshKey reports whether the server is configured with a mesh key.
 func (s *Server) HasMeshKey() bool { return s.meshKey != "" }
 
@@ -322,25 +343,40 @@ func (s *Server) initMetacert() {
 func (s *Server) MetaCert() []byte { return s.metaCert }
 
 // registerClient notes that client c is now authenticated and ready for packets.
-// If c's public key was already connected with a different connection, the prior one is closed.
-func (s *Server) registerClient(c *sclient) {
+//
+// If c's public key was already connected with a different
+// connection, the prior one is closed, unless it's fighting rapidly
+// with another client with the same key, in which case the returned
+// ok is false, and the caller should wait the provided duration
+// before trying again.
+func (s *Server) registerClient(c *sclient) (ok bool, d time.Duration) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 	old := s.clients[c.key]
 	if old == nil {
 		c.logf("adding connection")
 	} else {
+		// Take over the old rate limiter, discarding the one
+		// our caller just made.
+		c.replaceLimiter = old.replaceLimiter
+		if rr := c.replaceLimiter.ReserveN(timeNow(), 1); rr.OK() {
+			if d := rr.DelayFrom(timeNow()); d > 0 {
+				s.clientsReplaceLimited.Add(1)
+				return false, d
+			}
+		}
 		s.clientsReplaced.Add(1)
 		c.logf("adding connection, replacing %s", old.remoteAddr)
 		go old.nc.Close()
 	}
 	s.clients[c.key] = c
-	s.clientsEver[c.key] = true
 	if _, ok := s.clientsMesh[c.key]; !ok {
 		s.clientsMesh[c.key] = nil // just for varz of total users in cluster
 	}
+	s.keyOfAddr[c.remoteIPPort] = c.key
 	s.curClients.Add(1)
 	s.broadcastPeerStateChangeLocked(c.key, true)
+	return true, 0
 }
 
 // broadcastPeerStateChangeLocked enqueues a message to all watchers
@@ -373,6 +409,8 @@ func (s *Server) unregisterClient(c *sclient) {
 		delete(s.watchers, c)
 	}
 
+	delete(s.keyOfAddr, c.remoteIPPort)
+
 	s.curClients.Add(-1)
 	if c.preferred {
 		s.curHomeClients.Add(-1)
@@ -426,8 +464,9 @@ func (s *Server) addWatcher(c *sclient) {
 }
 
 func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connNum int64) error {
-	br, bw := brw.Reader, brw.Writer
+	br := brw.Reader
 	nc.SetDeadline(time.Now().Add(10 * time.Second))
+	bw := &lazyBufioWriter{w: nc, lbw: brw.Writer}
 	if err := s.sendServerKey(bw); err != nil {
 		return fmt.Errorf("send server key: %v", err)
 	}
@@ -446,21 +485,32 @@ func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connN
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
+	remoteIPPort, _ := netaddr.ParseIPPort(remoteAddr)
+
 	c := &sclient{
-		connNum:     connNum,
-		s:           s,
-		key:         clientKey,
-		nc:          nc,
-		br:          br,
-		bw:          bw,
-		logf:        logger.WithPrefix(s.logf, fmt.Sprintf("derp client %v/%x: ", remoteAddr, clientKey)),
-		done:        ctx.Done(),
-		remoteAddr:  remoteAddr,
-		connectedAt: time.Now(),
-		sendQueue:   make(chan pkt, perClientSendQueueDepth),
-		peerGone:    make(chan key.Public),
-		canMesh:     clientInfo.MeshKey != "" && clientInfo.MeshKey == s.meshKey,
+		connNum:        connNum,
+		s:              s,
+		key:            clientKey,
+		nc:             nc,
+		br:             br,
+		bw:             bw,
+		logf:           logger.WithPrefix(s.logf, fmt.Sprintf("derp client %v/%x: ", remoteAddr, clientKey)),
+		done:           ctx.Done(),
+		remoteAddr:     remoteAddr,
+		remoteIPPort:   remoteIPPort,
+		connectedAt:    time.Now(),
+		sendQueue:      make(chan pkt, perClientSendQueueDepth),
+		discoSendQueue: make(chan pkt, perClientSendQueueDepth),
+		peerGone:       make(chan key.Public),
+		canMesh:        clientInfo.MeshKey != "" && clientInfo.MeshKey == s.meshKey,
+
+		// Allow kicking out previous connections once a
+		// minute, with a very high burst of 100. Once a
+		// minute is less than the client's 2 minute
+		// inactivity timeout.
+		replaceLimiter: rate.NewLimiter(rate.Every(time.Minute), 100),
 	}
+
 	if c.canMesh {
 		c.meshUpdate = make(chan struct{})
 	}
@@ -468,10 +518,18 @@ func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connN
 		c.info = *clientInfo
 	}
 
-	s.registerClient(c)
+	for {
+		ok, d := s.registerClient(c)
+		if ok {
+			break
+		}
+		s.clientsReplaceSleeping.Add(1)
+		timeSleep(d)
+		s.clientsReplaceSleeping.Add(-1)
+	}
 	defer s.unregisterClient(c)
 
-	err = s.sendServerInfo(bw, clientKey)
+	err = s.sendServerInfo(c.bw, clientKey)
 	if err != nil {
 		return fmt.Errorf("send server info: %v", err)
 	}
@@ -479,6 +537,12 @@ func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connN
 	return c.run(ctx)
 }
 
+// for testing
+var (
+	timeSleep = time.Sleep
+	timeNow   = time.Now
+)
+
 // run serves the client until there's an error.
 // If the client hangs up or the server is closed, run returns nil, otherwise run returns an error.
 func (c *sclient) run(ctx context.Context) error {
@@ -602,17 +666,14 @@ func (c *sclient) handleFrameForwardPacket(ft frameType, fl uint32) error {
 	s.mu.Unlock()
 
 	if dst == nil {
-		s.packetsDropped.Add(1)
-		s.packetsDroppedFwdUnknown.Add(1)
-		if debug {
-			c.logf("dropping forwarded packet for unknown %x", dstKey)
-		}
+		s.recordDrop(contents, srcKey, dstKey, dropReasonUnknownDestOnFwd)
 		return nil
 	}
 
 	return c.sendPkt(dst, pkt{
-		bs:  contents,
-		src: srcKey,
+		bs:         contents,
+		enqueuedAt: time.Now(),
+		src:        srcKey,
 	})
 }
 
@@ -656,21 +717,53 @@ func (c *sclient) handleFrameSendPacket(ft frameType, fl uint32) error {
 			}
 			return nil
 		}
-		s.packetsDropped.Add(1)
-		s.packetsDroppedUnknown.Add(1)
-		if debug {
-			c.logf("dropping packet for unknown %x", dstKey)
-		}
+		s.recordDrop(contents, c.key, dstKey, dropReasonUnknownDest)
 		return nil
 	}
 
 	p := pkt{
-		bs:  contents,
-		src: c.key,
+		bs:         contents,
+		enqueuedAt: time.Now(),
+		src:        c.key,
 	}
 	return c.sendPkt(dst, p)
 }
 
+// dropReason is why we dropped a DERP frame.
+type dropReason int
+
+//go:generate go run tailscale.com/cmd/addlicense -year 2021 -file dropreason_string.go go run golang.org/x/tools/cmd/stringer -type=dropReason -trimprefix=dropReason
+
+const (
+	dropReasonUnknownDest      dropReason = iota // unknown destination pubkey
+	dropReasonUnknownDestOnFwd                   // unknown destination pubkey on a derp-forwarded packet
+	dropReasonGone                               // destination tailscaled disconnected before we could send
+	dropReasonQueueHead                          // destination queue is full, dropped packet at queue head
+	dropReasonQueueTail                          // destination queue is full, dropped packet at queue tail
+	dropReasonWriteError                         // OS write() failed
+)
+
+func (s *Server) recordDrop(packetBytes []byte, srcKey, dstKey key.Public, reason dropReason) {
+	s.packetsDropped.Add(1)
+	s.packetsDroppedReasonCounters[reason].Add(1)
+	if disco.LooksLikeDiscoWrapper(packetBytes) {
+		s.packetsDroppedTypeDisco.Add(1)
+	} else {
+		s.packetsDroppedTypeOther.Add(1)
+	}
+	if verboseDropKeys[dstKey] {
+		// Preformat the log string prior to calling limitedLogf. The
+		// limiter acts based on the format string, and we want to
+		// rate-limit per src/dst keys, not on the generic "dropped
+		// stuff" message.
+		msg := fmt.Sprintf("drop (%s) %s -> %s", srcKey.ShortString(), reason, dstKey.ShortString())
+		s.limitedLogf(msg)
+	}
+	if debug {
+		s.logf("dropping packet reason=%s dst=%s disco=%v", reason, dstKey, disco.LooksLikeDiscoWrapper(packetBytes))
+	}
+}
+
 func (c *sclient) sendPkt(dst *sclient, p pkt) error {
 	s := c.s
 	dstKey := dst.key
@@ -678,53 +771,34 @@ func (c *sclient) sendPkt(dst *sclient, p pkt) error {
 	// Attempt to queue for sending up to 3 times. On each attempt, if
 	// the queue is full, try to drop from queue head to prioritize
 	// fresher packets.
+	sendQueue := dst.sendQueue
+	if disco.LooksLikeDiscoWrapper(p.bs) {
+		sendQueue = dst.discoSendQueue
+	}
 	for attempt := 0; attempt < 3; attempt++ {
 		select {
 		case <-dst.done:
-			s.packetsDropped.Add(1)
-			s.packetsDroppedGone.Add(1)
-			if debug {
-				c.logf("dropping packet for shutdown client %x", dstKey)
-			}
+			s.recordDrop(p.bs, c.key, dstKey, dropReasonGone)
 			return nil
 		default:
 		}
 		select {
-		case dst.sendQueue <- p:
+		case sendQueue <- p:
 			return nil
 		default:
 		}
 
 		select {
-		case <-dst.sendQueue:
-			s.packetsDropped.Add(1)
-			s.packetsDroppedQueueHead.Add(1)
-			if verboseDropKeys[dstKey] {
-				// Generate a full string including src and dst, so
-				// the limiter kicks in once per src.
-				msg := fmt.Sprintf("tail drop %s -> %s", p.src.ShortString(), dstKey.ShortString())
-				c.s.limitedLogf(msg)
-			}
-			if debug {
-				c.logf("dropping packet from client %x queue head", dstKey)
-			}
+		case pkt := <-sendQueue:
+			s.recordDrop(pkt.bs, c.key, dstKey, dropReasonQueueHead)
+			c.recordQueueTime(pkt.enqueuedAt)
 		default:
 		}
 	}
 	// Failed to make room for packet. This can happen in a heavily
 	// contended queue with racing writers. Give up and tail-drop in
 	// this case to keep reader unblocked.
-	s.packetsDropped.Add(1)
-	s.packetsDroppedQueueTail.Add(1)
-	if verboseDropKeys[dstKey] {
-		// Generate a full string including src and dst, so
-		// the limiter kicks in once per src.
-		msg := fmt.Sprintf("head drop %s -> %s", p.src.ShortString(), dstKey.ShortString())
-		c.s.limitedLogf(msg)
-	}
-	if debug {
-		c.logf("dropping packet from client %x queue tail", dstKey)
-	}
+	s.recordDrop(p.bs, c.key, dstKey, dropReasonQueueTail)
 
 	return nil
 }
@@ -750,23 +824,37 @@ func (c *sclient) requestMeshUpdate() {
 }
 
 func (s *Server) verifyClient(clientKey key.Public, info *clientInfo) error {
-	// TODO(crawshaw): implement policy constraints on who can use the DERP server
-	// TODO(bradfitz): ... and at what rate.
+	if !s.verifyClients {
+		return nil
+	}
+	status, err := tailscale.Status(context.TODO())
+	if err != nil {
+		return fmt.Errorf("failed to query local tailscaled status: %w", err)
+	}
+	if clientKey == status.Self.PublicKey {
+		return nil
+	}
+	if _, exists := status.Peer[clientKey]; !exists {
+		return fmt.Errorf("client %v not in set of peers", clientKey)
+	}
+	// TODO(bradfitz): add policy for configurable bandwidth rate per client?
 	return nil
 }
 
-func (s *Server) sendServerKey(bw *bufio.Writer) error {
+func (s *Server) sendServerKey(lw *lazyBufioWriter) error {
 	buf := make([]byte, 0, len(magic)+len(s.publicKey))
 	buf = append(buf, magic...)
 	buf = append(buf, s.publicKey[:]...)
-	return writeFrame(bw, frameServerKey, buf)
+	err := writeFrame(lw.bw(), frameServerKey, buf)
+	lw.Flush() // redundant (no-op) flush to release bufio.Writer
+	return err
 }
 
 type serverInfo struct {
 	Version int `json:"version,omitempty"`
 }
 
-func (s *Server) sendServerInfo(bw *bufio.Writer, clientKey key.Public) error {
+func (s *Server) sendServerInfo(bw *lazyBufioWriter, clientKey key.Public) error {
 	var nonce [24]byte
 	if _, err := crand.Read(nonce[:]); err != nil {
 		return err
@@ -777,7 +865,7 @@ func (s *Server) sendServerInfo(bw *bufio.Writer, clientKey key.Public) error {
 	}
 
 	msgbox := box.Seal(nil, msg, &nonce, clientKey.B32(), s.privateKey.B32())
-	if err := writeFrameHeader(bw, frameServerInfo, nonceLen+uint32(len(msgbox))); err != nil {
+	if err := writeFrameHeader(bw.bw(), frameServerInfo, nonceLen+uint32(len(msgbox))); err != nil {
 		return err
 	}
 	if _, err := bw.Write(nonce[:]); err != nil {
@@ -885,18 +973,25 @@ func (s *Server) recvForwardPacket(br *bufio.Reader, frameLen uint32) (srcKey, d
 // (The "s" prefix is to more explicitly distinguish it from Client in derp_client.go)
 type sclient struct {
 	// Static after construction.
-	connNum    int64 // process-wide unique counter, incremented each Accept
-	s          *Server
-	nc         Conn
-	key        key.Public
-	info       clientInfo
-	logf       logger.Logf
-	done       <-chan struct{} // closed when connection closes
-	remoteAddr string          // usually ip:port from net.Conn.RemoteAddr().String()
-	sendQueue  chan pkt        // packets queued to this client; never closed
-	peerGone   chan key.Public // write request that a previous sender has disconnected (not used by mesh peers)
-	meshUpdate chan struct{}   // write request to write peerStateChange
-	canMesh    bool            // clientInfo had correct mesh token for inter-region routing
+	connNum        int64 // process-wide unique counter, incremented each Accept
+	s              *Server
+	nc             Conn
+	key            key.Public
+	info           clientInfo
+	logf           logger.Logf
+	done           <-chan struct{} // closed when connection closes
+	remoteAddr     string          // usually ip:port from net.Conn.RemoteAddr().String()
+	remoteIPPort   netaddr.IPPort  // zero if remoteAddr is not ip:port.
+	sendQueue      chan pkt        // packets queued to this client; never closed
+	discoSendQueue chan pkt        // important packets queued to this client; never closed
+	peerGone       chan key.Public // write request that a previous sender has disconnected (not used by mesh peers)
+	meshUpdate     chan struct{}   // write request to write peerStateChange
+	canMesh        bool            // clientInfo had correct mesh token for inter-region routing
+
+	// replaceLimiter controls how quickly two connections with
+	// the same client key can kick each other off the server by
+	// taking over ownership of a key.
+	replaceLimiter *rate.Limiter
 
 	// Owned by run, not thread-safe.
 	br          *bufio.Reader
@@ -904,7 +999,7 @@ type sclient struct {
 	preferred   bool
 
 	// Owned by sender, not thread-safe.
-	bw *bufio.Writer
+	bw *lazyBufioWriter
 
 	// Guarded by s.mu
 	//
@@ -927,11 +1022,13 @@ type pkt struct {
 	// src is the who's the sender of the packet.
 	src key.Public
 
+	// enqueuedAt is when a packet was put onto a queue before it was sent,
+	// and is used for reporting metrics on the duration of packets in the queue.
+	enqueuedAt time.Time
+
 	// bs is the data packet bytes.
 	// The memory is owned by pkt.
 	bs []byte
-
-	// TODO(danderson): enqueue time, to measure queue latency?
 }
 
 func (c *sclient) setPreferred(v bool) {
@@ -959,6 +1056,25 @@ func (c *sclient) setPreferred(v bool) {
 	}
 }
 
+// expMovingAverage returns the new moving average given the previous average,
+// a new value, and an alpha decay factor.
+// https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average
+func expMovingAverage(prev, newValue, alpha float64) float64 {
+	return alpha*newValue + (1-alpha)*prev
+}
+
+// recordQueueTime updates the average queue duration metric after a packet has been sent.
+func (c *sclient) recordQueueTime(enqueuedAt time.Time) {
+	elapsed := float64(time.Since(enqueuedAt).Milliseconds())
+	for {
+		old := atomic.LoadUint64(c.s.avgQueueDuration)
+		newAvg := expMovingAverage(math.Float64frombits(old), elapsed, 0.1)
+		if atomic.CompareAndSwapUint64(c.s.avgQueueDuration, old, math.Float64bits(newAvg)) {
+			break
+		}
+	}
+}
+
 func (c *sclient) sendLoop(ctx context.Context) error {
 	defer func() {
 		// If the sender shuts down unilaterally due to an error, close so
@@ -968,12 +1084,10 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 		// Drain the send queue to count dropped packets
 		for {
 			select {
-			case <-c.sendQueue:
-				c.s.packetsDropped.Add(1)
-				c.s.packetsDroppedGone.Add(1)
-				if debug {
-					c.logf("dropping packet for shutdown %x", c.key)
-				}
+			case pkt := <-c.sendQueue:
+				c.s.recordDrop(pkt.bs, pkt.src, c.key, dropReasonGone)
+			case pkt := <-c.discoSendQueue:
+				c.s.recordDrop(pkt.bs, pkt.src, c.key, dropReasonGone)
 			default:
 				return
 			}
@@ -1002,6 +1116,11 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 			continue
 		case msg := <-c.sendQueue:
 			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
+			continue
+		case msg := <-c.discoSendQueue:
+			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
 			continue
 		case <-keepAliveTick.C:
 			werr = c.sendKeepAlive()
@@ -1025,6 +1144,10 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 			continue
 		case msg := <-c.sendQueue:
 			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
+		case msg := <-c.discoSendQueue:
+			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
 		case <-keepAliveTick.C:
 			werr = c.sendKeepAlive()
 		}
@@ -1038,14 +1161,14 @@ func (c *sclient) setWriteDeadline() {
 // sendKeepAlive sends a keep-alive frame, without flushing.
 func (c *sclient) sendKeepAlive() error {
 	c.setWriteDeadline()
-	return writeFrameHeader(c.bw, frameKeepAlive, 0)
+	return writeFrameHeader(c.bw.bw(), frameKeepAlive, 0)
 }
 
 // sendPeerGone sends a peerGone frame, without flushing.
 func (c *sclient) sendPeerGone(peer key.Public) error {
 	c.s.peerGoneFrames.Add(1)
 	c.setWriteDeadline()
-	if err := writeFrameHeader(c.bw, framePeerGone, keyLen); err != nil {
+	if err := writeFrameHeader(c.bw.bw(), framePeerGone, keyLen); err != nil {
 		return err
 	}
 	_, err := c.bw.Write(peer[:])
@@ -1055,7 +1178,7 @@ func (c *sclient) sendPeerGone(peer key.Public) error {
 // sendPeerPresent sends a peerPresent frame, without flushing.
 func (c *sclient) sendPeerPresent(peer key.Public) error {
 	c.setWriteDeadline()
-	if err := writeFrameHeader(c.bw, framePeerPresent, keyLen); err != nil {
+	if err := writeFrameHeader(c.bw.bw(), framePeerPresent, keyLen); err != nil {
 		return err
 	}
 	_, err := c.bw.Write(peer[:])
@@ -1114,11 +1237,7 @@ func (c *sclient) sendPacket(srcKey key.Public, contents []byte) (err error) {
 	defer func() {
 		// Stats update.
 		if err != nil {
-			c.s.packetsDropped.Add(1)
-			c.s.packetsDroppedWrite.Add(1)
-			if debug {
-				c.logf("dropping packet to %x: %v", c.key, err)
-			}
+			c.s.recordDrop(contents, srcKey, c.key, dropReasonWriteError)
 		} else {
 			c.s.packetsSent.Add(1)
 			c.s.bytesSent.Add(int64(len(contents)))
@@ -1132,11 +1251,11 @@ func (c *sclient) sendPacket(srcKey key.Public, contents []byte) (err error) {
 	if withKey {
 		pktLen += len(srcKey)
 	}
-	if err = writeFrameHeader(c.bw, frameRecvPacket, uint32(pktLen)); err != nil {
+	if err = writeFrameHeader(c.bw.bw(), frameRecvPacket, uint32(pktLen)); err != nil {
 		return err
 	}
 	if withKey {
-		err := writePublicKey(c.bw, &srcKey)
+		err := writePublicKey(c.bw.bw(), &srcKey)
 		if err != nil {
 			return err
 		}
@@ -1264,7 +1383,6 @@ func (s *Server) expVarFunc(f func() interface{}) expvar.Func {
 // ExpVar returns an expvar variable suitable for registering with expvar.Publish.
 func (s *Server) ExpVar() expvar.Var {
 	m := new(metrics.Set)
-	m.Set("counter_unique_clients_ever", s.expVarFunc(func() interface{} { return len(s.clientsEver) }))
 	m.Set("gauge_memstats_sys0", expvar.Func(func() interface{} { return int64(s.memSys0) }))
 	m.Set("gauge_watchers", s.expVarFunc(func() interface{} { return len(s.watchers) }))
 	m.Set("gauge_current_connections", &s.curClients)
@@ -1274,10 +1392,13 @@ func (s *Server) ExpVar() expvar.Var {
 	m.Set("gauge_clients_remote", expvar.Func(func() interface{} { return len(s.clientsMesh) - len(s.clients) }))
 	m.Set("accepts", &s.accepts)
 	m.Set("clients_replaced", &s.clientsReplaced)
+	m.Set("clients_replace_limited", &s.clientsReplaceLimited)
+	m.Set("gauge_clients_replace_sleeping", &s.clientsReplaceSleeping)
 	m.Set("bytes_received", &s.bytesRecv)
 	m.Set("bytes_sent", &s.bytesSent)
 	m.Set("packets_dropped", &s.packetsDropped)
 	m.Set("counter_packets_dropped_reason", &s.packetsDroppedReason)
+	m.Set("counter_packets_dropped_type", &s.packetsDroppedType)
 	m.Set("counter_packets_received_kind", &s.packetsRecvByKind)
 	m.Set("packets_sent", &s.packetsSent)
 	m.Set("packets_received", &s.packetsRecv)
@@ -1290,6 +1411,9 @@ func (s *Server) ExpVar() expvar.Var {
 	m.Set("multiforwarder_created", &s.multiForwarderCreated)
 	m.Set("multiforwarder_deleted", &s.multiForwarderDeleted)
 	m.Set("packet_forwarder_delete_other_value", &s.removePktForwardOther)
+	m.Set("average_queue_duration_ms", expvar.Func(func() interface{} {
+		return math.Float64frombits(atomic.LoadUint64(s.avgQueueDuration))
+	}))
 	var expvarVersion expvar.String
 	expvarVersion.Set(version.Long)
 	m.Set("version", &expvarVersion)
@@ -1365,3 +1489,126 @@ func writePublicKey(bw *bufio.Writer, key *key.Public) error {
 	}
 	return nil
 }
+
+const minTimeBetweenLogs = 2 * time.Second
+
+// BytesSentRecv records the number of bytes that have been sent since the last traffic check
+// for a given process, as well as the public key of the process sending those bytes.
+type BytesSentRecv struct {
+	Sent uint64
+	Recv uint64
+	// Key is the public key of the client which sent/received these bytes.
+	Key key.Public
+}
+
+// parseSSOutput parses the output from the specific call to ss in ServeDebugTraffic.
+// Separated out for ease of testing.
+func parseSSOutput(raw string) map[netaddr.IPPort]BytesSentRecv {
+	newState := map[netaddr.IPPort]BytesSentRecv{}
+	// parse every 2 lines and get src and dst ips, and kv pairs
+	lines := strings.Split(raw, "\n")
+	for i := 0; i < len(lines); i += 2 {
+		ipInfo := strings.Fields(strings.TrimSpace(lines[i]))
+		if len(ipInfo) < 5 {
+			continue
+		}
+		src, err := netaddr.ParseIPPort(ipInfo[4])
+		if err != nil {
+			continue
+		}
+		stats := strings.Fields(strings.TrimSpace(lines[i+1]))
+		stat := BytesSentRecv{}
+		for _, s := range stats {
+			if strings.Contains(s, "bytes_sent") {
+				sent, err := strconv.Atoi(s[strings.Index(s, ":")+1:])
+				if err == nil {
+					stat.Sent = uint64(sent)
+				}
+			} else if strings.Contains(s, "bytes_received") {
+				recv, err := strconv.Atoi(s[strings.Index(s, ":")+1:])
+				if err == nil {
+					stat.Recv = uint64(recv)
+				}
+			}
+		}
+		newState[src] = stat
+	}
+	return newState
+}
+
+func (s *Server) ServeDebugTraffic(w http.ResponseWriter, r *http.Request) {
+	prevState := map[netaddr.IPPort]BytesSentRecv{}
+	enc := json.NewEncoder(w)
+	for r.Context().Err() == nil {
+		output, err := exec.Command("ss", "-i", "-H", "-t").Output()
+		if err != nil {
+			fmt.Fprintf(w, "ss failed: %v", err)
+			return
+		}
+		newState := parseSSOutput(string(output))
+		s.mu.Lock()
+		for k, next := range newState {
+			prev := prevState[k]
+			if prev.Sent < next.Sent || prev.Recv < next.Recv {
+				if pkey, ok := s.keyOfAddr[k]; ok {
+					next.Key = pkey
+					if err := enc.Encode(next); err != nil {
+						s.mu.Unlock()
+						return
+					}
+				}
+			}
+		}
+		s.mu.Unlock()
+		prevState = newState
+		if _, err := fmt.Fprintln(w); err != nil {
+			return
+		}
+		if f, ok := w.(http.Flusher); ok {
+			f.Flush()
+		}
+		time.Sleep(minTimeBetweenLogs)
+	}
+}
+
+var bufioWriterPool = &sync.Pool{
+	New: func() interface{} {
+		return bufio.NewWriterSize(ioutil.Discard, 2<<10)
+	},
+}
+
+// lazyBufioWriter is a bufio.Writer-like wrapping writer that lazily
+// allocates its actual bufio.Writer from a sync.Pool, releasing it to
+// the pool upon flush.
+//
+// We do this to reduce memory overhead; most DERP connections are
+// idle and the idle bufio.Writers were 30% of overall memory usage.
+type lazyBufioWriter struct {
+	w   io.Writer     // underlying
+	lbw *bufio.Writer // lazy; nil means it needs an associated buffer
+}
+
+func (w *lazyBufioWriter) bw() *bufio.Writer {
+	if w.lbw == nil {
+		w.lbw = bufioWriterPool.Get().(*bufio.Writer)
+		w.lbw.Reset(w.w)
+	}
+	return w.lbw
+}
+
+func (w *lazyBufioWriter) Available() int { return w.bw().Available() }
+
+func (w *lazyBufioWriter) Write(p []byte) (int, error) { return w.bw().Write(p) }
+
+func (w *lazyBufioWriter) Flush() error {
+	if w.lbw == nil {
+		return nil
+	}
+	err := w.lbw.Flush()
+
+	w.lbw.Reset(ioutil.Discard)
+	bufioWriterPool.Put(w.lbw)
+	w.lbw = nil
+
+	return err
+}

derp/derp_test.go

@@ -23,6 +23,7 @@ import (
 	"testing"
 	"time"
 
+	"golang.org/x/time/rate"
 	"tailscale.com/net/nettest"
 	"tailscale.com/types/key"
 	"tailscale.com/types/logger"
@@ -409,7 +410,7 @@ func TestSendFreeze(t *testing.T) {
 	for i := 0; i < cap(errCh); i++ {
 		err := <-errCh
 		if err != nil {
-			if errors.Is(err, io.EOF) {
+			if errors.Is(err, io.EOF) || errors.Is(err, net.ErrClosed) {
 				continue
 			}
 			t.Error(err)
@@ -849,6 +850,136 @@ func TestClientSendPong(t *testing.T) {
 
 }
 
+func TestServerReplaceClients(t *testing.T) {
+	defer func() {
+		timeSleep = time.Sleep
+		timeNow = time.Now
+	}()
+
+	var (
+		mu     sync.Mutex
+		now    = time.Unix(123, 0)
+		sleeps int
+		slept  time.Duration
+	)
+	timeSleep = func(d time.Duration) {
+		mu.Lock()
+		defer mu.Unlock()
+		sleeps++
+		slept += d
+		now = now.Add(d)
+	}
+	timeNow = func() time.Time {
+		mu.Lock()
+		defer mu.Unlock()
+		return now
+	}
+
+	serverPrivateKey := newPrivateKey(t)
+	var logger logger.Logf = logger.Discard
+	const debug = false
+	if debug {
+		logger = t.Logf
+	}
+
+	s := NewServer(serverPrivateKey, logger)
+	defer s.Close()
+
+	priv := newPrivateKey(t)
+
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+
+	connNum := 0
+	connect := func() *Client {
+		connNum++
+		cout, err := net.Dial("tcp", ln.Addr().String())
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		cin, err := ln.Accept()
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		brwServer := bufio.NewReadWriter(bufio.NewReader(cin), bufio.NewWriter(cin))
+		go s.Accept(cin, brwServer, fmt.Sprintf("test-client-%d", connNum))
+
+		brw := bufio.NewReadWriter(bufio.NewReader(cout), bufio.NewWriter(cout))
+		c, err := NewClient(priv, cout, brw, logger)
+		if err != nil {
+			t.Fatalf("client %d: %v", connNum, err)
+		}
+		return c
+	}
+
+	wantVar := func(v *expvar.Int, want int64) {
+		t.Helper()
+		if got := v.Value(); got != want {
+			t.Errorf("got %d; want %d", got, want)
+		}
+	}
+
+	wantClosed := func(c *Client) {
+		t.Helper()
+		for {
+			m, err := c.Recv()
+			if err != nil {
+				t.Logf("got expected error: %v", err)
+				return
+			}
+			switch m.(type) {
+			case ServerInfoMessage:
+				continue
+			default:
+				t.Fatalf("client got %T; wanted an error", m)
+			}
+		}
+	}
+
+	c1 := connect()
+	waitConnect(t, c1)
+	c2 := connect()
+	waitConnect(t, c2)
+	wantVar(&s.clientsReplaced, 1)
+	wantClosed(c1)
+
+	for i := 0; i < 100+5; i++ {
+		c := connect()
+		defer c.nc.Close()
+		if s.clientsReplaceLimited.Value() == 0 && i < 90 {
+			continue
+		}
+		t.Logf("for %d: replaced=%d, limited=%d, sleeping=%d", i,
+			s.clientsReplaced.Value(),
+			s.clientsReplaceLimited.Value(),
+			s.clientsReplaceSleeping.Value(),
+		)
+	}
+
+	mu.Lock()
+	defer mu.Unlock()
+	if sleeps == 0 {
+		t.Errorf("no sleeps")
+	}
+	if slept == 0 {
+		t.Errorf("total sleep duration was 0")
+	}
+}
+
+func TestLimiter(t *testing.T) {
+	rl := rate.NewLimiter(rate.Every(time.Minute), 100)
+	for i := 0; i < 200; i++ {
+		r := rl.Reserve()
+		d := r.Delay()
+		t.Logf("i=%d, allow=%v, d=%v", i, r.OK(), d)
+	}
+}
+
 func BenchmarkSendRecv(b *testing.B) {
 	for _, size := range []int{10, 100, 1000, 10000} {
 		b.Run(fmt.Sprintf("msgsize=%d", size), func(b *testing.B) { benchmarkSendRecvSize(b, size) })
@@ -948,3 +1079,14 @@ func waitConnect(t testing.TB, c *Client) {
 		t.Fatalf("client first Recv was unexpected type %T", v)
 	}
 }
+
+func TestParseSSOutput(t *testing.T) {
+	contents, err := ioutil.ReadFile("testdata/example_ss.txt")
+	if err != nil {
+		t.Errorf("ioutil.Readfile(example_ss.txt) failed: %v", err)
+	}
+	seen := parseSSOutput(string(contents))
+	if len(seen) == 0 {
+		t.Errorf("parseSSOutput expected non-empty map")
+	}
+}

derp/derphttp/derphttp_client.go

@@ -50,9 +50,11 @@ type Client struct {
 	TLSConfig *tls.Config        // optional; nil means default
 	DNSCache  *dnscache.Resolver // optional; nil means no caching
 	MeshKey   string             // optional; for trusted clients
+	IsProber  bool               // optional; for probers to optional declare themselves as such
 
 	privateKey key.Private
 	logf       logger.Logf
+	dialer     func(ctx context.Context, network, addr string) (net.Conn, error)
 
 	// Either url or getRegion is non-nil:
 	url       *url.URL
@@ -130,6 +132,11 @@ func (c *Client) ServerPublicKey() key.Public {
 	return c.serverPubKey
 }
 
+// SelfPublicKey returns our own public key.
+func (c *Client) SelfPublicKey() key.Public {
+	return c.privateKey.Public()
+}
+
 func urlPort(u *url.URL) string {
 	if p := u.Port(); p != "" {
 		return p
@@ -338,6 +345,7 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 		derp.MeshKey(c.MeshKey),
 		derp.ServerPublicKey(serverPub),
 		derp.CanAckPings(c.canAckPings),
+		derp.IsProber(c.IsProber),
 	)
 	if err != nil {
 		return nil, 0, err
@@ -356,14 +364,26 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 	return c.client, c.connGen, nil
 }
 
+// SetURLDialer sets the dialer to use for dialing URLs.
+// This dialer is only use for clients created with NewClient, not NewRegionClient.
+// If unset or nil, the default dialer is used.
+//
+// The primary use for this is the derper mesh mode to connect to each
+// other over a VPC network.
+func (c *Client) SetURLDialer(dialer func(ctx context.Context, network, addr string) (net.Conn, error)) {
+	c.dialer = dialer
+}
+
 func (c *Client) dialURL(ctx context.Context) (net.Conn, error) {
 	host := c.url.Hostname()
+	if c.dialer != nil {
+		return c.dialer(ctx, "tcp", net.JoinHostPort(host, urlPort(c.url)))
+	}
 	hostOrIP := host
-
 	dialer := netns.NewDialer()
 
 	if c.DNSCache != nil {
-		ip, _, err := c.DNSCache.LookupIP(ctx, host)
+		ip, _, _, err := c.DNSCache.LookupIP(ctx, host)
 		if err == nil {
 			hostOrIP = ip.String()
 		}
@@ -410,9 +430,7 @@ func (c *Client) dialRegion(ctx context.Context, reg *tailcfg.DERPRegion) (net.C
 func (c *Client) tlsClient(nc net.Conn, node *tailcfg.DERPNode) *tls.Conn {
 	tlsConf := tlsdial.Config(c.tlsServerName(node), c.TLSConfig)
 	if node != nil {
-		if node.DERPTestPort != 0 {
-			tlsConf.InsecureSkipVerify = true
-		}
+		tlsConf.InsecureSkipVerify = node.InsecureForTests
 		if node.CertName != "" {
 			tlsdial.SetConfigExpectedCert(tlsConf, node.CertName)
 		}
@@ -511,8 +529,8 @@ func (c *Client) dialNode(ctx context.Context, n *tailcfg.DERPNode) (net.Conn, e
 				dst = n.HostName
 			}
 			port := "443"
-			if n.DERPTestPort != 0 {
-				port = fmt.Sprint(n.DERPTestPort)
+			if n.DERPPort != 0 {
+				port = fmt.Sprint(n.DERPPort)
 			}
 			c, err := c.dialContext(ctx, proto, net.JoinHostPort(dst, port))
 			select {

derp/derpmap/derpmap.go

@@ -1,88 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package derpmap contains information about Tailscale.com's production DERP nodes.
-//
-// This package is only used by the "tailscale netcheck" command for debugging.
-// In normal operation the Tailscale nodes get this sent to them from the control
-// server.
-//
-// TODO: remove this package and make "tailscale netcheck" get the
-// list from the control server too.
-package derpmap
-
-import (
-	"fmt"
-	"strings"
-
-	"tailscale.com/tailcfg"
-)
-
-func derpNode(suffix, v4, v6 string) *tailcfg.DERPNode {
-	return &tailcfg.DERPNode{
-		Name:     suffix, // updated later
-		RegionID: 0,      // updated later
-		IPv4:     v4,
-		IPv6:     v6,
-	}
-}
-
-func derpRegion(id int, code, name string, nodes ...*tailcfg.DERPNode) *tailcfg.DERPRegion {
-	region := &tailcfg.DERPRegion{
-		RegionID:   id,
-		RegionName: name,
-		RegionCode: code,
-		Nodes:      nodes,
-	}
-	for _, n := range nodes {
-		n.Name = fmt.Sprintf("%d%s", id, n.Name)
-		n.RegionID = id
-		n.HostName = fmt.Sprintf("derp%s.tailscale.com", strings.TrimSuffix(n.Name, "a"))
-	}
-	return region
-}
-
-// Prod returns Tailscale's map of relay servers.
-//
-// This list is only used by cmd/tailscale's netcheck subcommand. In
-// normal operation the Tailscale nodes get this sent to them from the
-// control server.
-//
-// This list is subject to change and should not be relied on.
-func Prod() *tailcfg.DERPMap {
-	return &tailcfg.DERPMap{
-		Regions: map[int]*tailcfg.DERPRegion{
-			1: derpRegion(1, "nyc", "New York City",
-				derpNode("a", "159.89.225.99", "2604:a880:400:d1::828:b001"),
-			),
-			2: derpRegion(2, "sfo", "San Francisco",
-				derpNode("a", "167.172.206.31", "2604:a880:2:d1::c5:7001"),
-			),
-			3: derpRegion(3, "sin", "Singapore",
-				derpNode("a", "68.183.179.66", "2400:6180:0:d1::67d:8001"),
-			),
-			4: derpRegion(4, "fra", "Frankfurt",
-				derpNode("a", "167.172.182.26", "2a03:b0c0:3:e0::36e:9001"),
-			),
-			5: derpRegion(5, "syd", "Sydney",
-				derpNode("a", "103.43.75.49", "2001:19f0:5801:10b7:5400:2ff:feaa:284c"),
-			),
-			6: derpRegion(6, "blr", "Bangalore",
-				derpNode("a", "68.183.90.120", "2400:6180:100:d0::982:d001"),
-			),
-			7: derpRegion(7, "tok", "Tokyo",
-				derpNode("a", "167.179.89.145", "2401:c080:1000:467f:5400:2ff:feee:22aa"),
-			),
-			8: derpRegion(8, "lhr", "London",
-				derpNode("a", "167.71.139.179", "2a03:b0c0:1:e0::3cc:e001"),
-			),
-			9: derpRegion(9, "dfw", "Dallas",
-				derpNode("a", "207.148.3.137", "2001:19f0:6401:1d9c:5400:2ff:feef:bb82"),
-			),
-			10: derpRegion(10, "sea", "Seattle",
-				derpNode("a", "137.220.36.168", "2001:19f0:8001:2d9:5400:2ff:feef:bbb1"),
-			),
-		},
-	}
-}

derp/dropreason_string.go

@@ -0,0 +1,32 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Code generated by "stringer -type=dropReason -trimprefix=dropReason"; DO NOT EDIT.
+
+package derp
+
+import "strconv"
+
+func _() {
+	// An "invalid array index" compiler error signifies that the constant values have changed.
+	// Re-run the stringer command to generate them again.
+	var x [1]struct{}
+	_ = x[dropReasonUnknownDest-0]
+	_ = x[dropReasonUnknownDestOnFwd-1]
+	_ = x[dropReasonGone-2]
+	_ = x[dropReasonQueueHead-3]
+	_ = x[dropReasonQueueTail-4]
+	_ = x[dropReasonWriteError-5]
+}
+
+const _dropReason_name = "UnknownDestUnknownDestOnFwdGoneQueueHeadQueueTailWriteError"
+
+var _dropReason_index = [...]uint8{0, 11, 27, 31, 40, 49, 59}
+
+func (i dropReason) String() string {
+	if i < 0 || i >= dropReason(len(_dropReason_index)-1) {
+		return "dropReason(" + strconv.FormatInt(int64(i), 10) + ")"
+	}
+	return _dropReason_name[_dropReason_index[i]:_dropReason_index[i+1]]
+}

derp/testdata/example_ss.txt

@@ -0,0 +1,8 @@
+ESTAB 0      0       10.255.1.11:35238  34.210.105.16:https
+         cubic wscale:7,7 rto:236 rtt:34.14/3.432 ato:40 mss:1448 pmtu:1500 rcvmss:1448 advmss:1448 cwnd:8 ssthresh:6 bytes_sent:38056577 bytes_retrans:2918 bytes_acked:38053660 bytes_received:6973211 segs_out:165090 segs_in:124227 data_segs_out:78018 data_segs_in:71645 send 2.71Mbps lastsnd:1156 lastrcv:1120 lastack:1120 pacing_rate 3.26Mbps delivery_rate 2.35Mbps delivered:78017 app_limited busy:2586132ms retrans:0/6 dsack_dups:4 reordering:5 reord_seen:15 rcv_rtt:126355 rcv_space:65780 rcv_ssthresh:541928 minrtt:26.632
+ESTAB 0      80     100.79.58.14:ssh    100.95.73.104:58145
+         cubic wscale:6,7 rto:224 rtt:23.051/2.03 ato:172 mss:1228 pmtu:1280 rcvmss:1228 advmss:1228 cwnd:10 ssthresh:94 bytes_sent:1591815 bytes_retrans:944 bytes_acked:1590791 bytes_received:158925 segs_out:8070 segs_in:8858 data_segs_out:7452 data_segs_in:3789 send 4.26Mbps lastsnd:4 lastrcv:4 lastack:4 pacing_rate 8.52Mbps delivery_rate 10.9Mbps delivered:7451 app_limited busy:61656ms unacked:2 retrans:0/10 dsack_dups:10 rcv_rtt:174712 rcv_space:65025 rcv_ssthresh:64296 minrtt:16.186                                   
+ESTAB 0      374     10.255.1.11:43254 167.172.206.31:https
+         cubic wscale:7,7 rto:224 rtt:22.55/1.941 ato:40 mss:1448 pmtu:1500 rcvmss:1448 advmss:1448 cwnd:6 ssthresh:4 bytes_sent:14594668 bytes_retrans:173314 bytes_acked:14420981 bytes_received:4207111 segs_out:80566 segs_in:70310 data_segs_out:24317 data_segs_in:20365 send 3.08Mbps lastsnd:4 lastrcv:4 lastack:4 pacing_rate 3.7Mbps delivery_rate 3.05Mbps delivered:24111 app_limited busy:184820ms unacked:2 retrans:0/185 dsack_dups:1 reord_seen:3 rcv_rtt:651.262 rcv_space:226657 rcv_ssthresh:1557136 minrtt:10.18           
+ESTAB 0      0       10.255.1.11:33036    3.121.18.47:https
+         cubic wscale:7,7 rto:372 rtt:168.408/2.044 ato:40 mss:1448 pmtu:1500 rcvmss:1448 advmss:1448 cwnd:10 bytes_sent:27500 bytes_acked:27501 bytes_received:1386524 segs_out:10990 segs_in:11037 data_segs_out:303 data_segs_in:3414 send 688kbps lastsnd:125776 lastrcv:9640 lastack:22760 pacing_rate 1.38Mbps delivery_rate 482kbps delivered:304 app_limited busy:43024ms rcv_rtt:3345.12 rcv_space:62431 rcv_ssthresh:760472 minrtt:168.867

disco/disco.go

@@ -147,9 +147,9 @@ const epLength = 16 + 2 // 16 byte IP address + 2 byte port
 func (m *CallMeMaybe) AppendMarshal(b []byte) []byte {
 	ret, p := appendMsgHeader(b, TypeCallMeMaybe, v0, epLength*len(m.MyNumber))
 	for _, ipp := range m.MyNumber {
-		a := ipp.IP.As16()
+		a := ipp.IP().As16()
 		copy(p[:], a[:])
-		binary.BigEndian.PutUint16(p[16:], ipp.Port)
+		binary.BigEndian.PutUint16(p[16:], ipp.Port())
 		p = p[epLength:]
 	}
 	return ret
@@ -164,10 +164,9 @@ func parseCallMeMaybe(ver uint8, p []byte) (m *CallMeMaybe, err error) {
 	for len(p) > 0 {
 		var a [16]byte
 		copy(a[:], p)
-		m.MyNumber = append(m.MyNumber, netaddr.IPPort{
-			IP:   netaddr.IPFrom16(a),
-			Port: binary.BigEndian.Uint16(p[16:18]),
-		})
+		m.MyNumber = append(m.MyNumber, netaddr.IPPortFrom(
+			netaddr.IPFrom16(a),
+			binary.BigEndian.Uint16(p[16:18])))
 		p = p[epLength:]
 	}
 	return m, nil
@@ -187,9 +186,9 @@ const pongLen = 12 + 16 + 2
 func (m *Pong) AppendMarshal(b []byte) []byte {
 	ret, d := appendMsgHeader(b, TypePong, v0, pongLen)
 	d = d[copy(d, m.TxID[:]):]
-	ip16 := m.Src.IP.As16()
+	ip16 := m.Src.IP().As16()
 	d = d[copy(d, ip16[:]):]
-	binary.BigEndian.PutUint16(d, m.Src.Port)
+	binary.BigEndian.PutUint16(d, m.Src.Port())
 	return ret
 }
 
@@ -201,10 +200,10 @@ func parsePong(ver uint8, p []byte) (m *Pong, err error) {
 	copy(m.TxID[:], p)
 	p = p[12:]
 
-	m.Src.IP, _ = netaddr.FromStdIP(net.IP(p[:16]))
+	srcIP, _ := netaddr.FromStdIP(net.IP(p[:16]))
 	p = p[16:]
-
-	m.Src.Port = binary.BigEndian.Uint16(p)
+	port := binary.BigEndian.Uint16(p)
+	m.Src = netaddr.IPPortFrom(srcIP, port)
 	return m, nil
 }
 

disco/disco_fuzzer.go

@@ -0,0 +1,19 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+//go:build gofuzz
+// +build gofuzz
+
+package disco
+
+func Fuzz(data []byte) int {
+	m, _ := Parse(data)
+
+	newBytes := m.AppendMarshal(data)
+	parsedMarshall, _ := Parse(newBytes)
+
+	if m != parsedMarshall {
+		panic("Parsing error")
+	}
+	return 1
+}

go.mod

@@ -3,45 +3,54 @@ module tailscale.com
 go 1.16
 
 require (
-	github.com/alexbrainman/sspi v0.0.0-20180613141037-e580b900e9f5
-	github.com/anmitsu/go-shlex v0.0.0-20161002113705-648efa622239 // indirect
-	github.com/coreos/go-iptables v0.4.5
-	github.com/flynn/go-shlex v0.0.0-20150515145356-3f9db97f8568 // indirect
-	github.com/github/certstore v0.1.0
-	github.com/gliderlabs/ssh v0.2.2
+	github.com/alexbrainman/sspi v0.0.0-20210105120005-909beea2cc74
+	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be // indirect
+	github.com/aws/aws-sdk-go v1.38.52
+	github.com/coreos/go-iptables v0.6.0
+	github.com/creack/pty v1.1.9
+	github.com/dave/jennifer v1.4.1
+	github.com/dsnet/golib/jsonfmt v1.0.0
+	github.com/frankban/quicktest v1.13.0
+	github.com/gliderlabs/ssh v0.3.2
 	github.com/go-multierror/multierror v1.0.2
-	github.com/go-ole/go-ole v1.2.4
-	github.com/godbus/dbus/v5 v5.0.3
-	github.com/google/go-cmp v0.5.4
-	github.com/goreleaser/nfpm v1.1.10
-	github.com/jsimonetti/rtnetlink v0.0.0-20210212075122-66c871082f2b
-	github.com/klauspost/compress v1.10.10
-	github.com/kr/pty v1.1.8
-	github.com/mdlayher/netlink v1.3.2
-	github.com/mdlayher/sdnotify v0.0.0-20200625151349-e4a4f32afc4a
-	github.com/miekg/dns v1.1.30
-	github.com/pborman/getopt v0.0.0-20190409184431-ee0cd42419d3
+	github.com/go-ole/go-ole v1.2.5
+	github.com/godbus/dbus/v5 v5.0.4
+	github.com/google/go-cmp v0.5.6
+	github.com/google/goexpect v0.0.0-20210430020637-ab937bf7fd6f
+	github.com/google/uuid v1.1.2
+	github.com/goreleaser/nfpm v1.10.3
+	github.com/iancoleman/strcase v0.2.0
+	github.com/insomniacslk/dhcp v0.0.0-20210621130208-1cac67f12b1e
+	github.com/jsimonetti/rtnetlink v0.0.0-20210525051524-4cc836578190
+	github.com/kballard/go-shellquote v0.0.0-20180428030007-95032a82bc51
+	github.com/klauspost/compress v1.12.2
+	github.com/mdlayher/netlink v1.4.1
+	github.com/mdlayher/sdnotify v0.0.0-20210228150836-ea3ec207d697
+	github.com/miekg/dns v1.1.42
+	github.com/mitchellh/go-ps v1.0.0
+	github.com/pborman/getopt v1.1.0
 	github.com/peterbourgon/ff/v2 v2.0.0
-	github.com/pkg/errors v0.9.1 // indirect
+	github.com/pkg/sftp v1.13.0
+	github.com/tailscale/certstore v0.0.0-20210528134328-066c94b793d3
 	github.com/tailscale/depaware v0.0.0-20201214215404-77d1e9757027
-	github.com/tailscale/wireguard-go v0.0.0-20210403171604-17614717a9b5
+	github.com/tailscale/goupnp v1.0.1-0.20210804011211-c64d0f06ea05
+	github.com/tailscale/hujson v0.0.0-20200924210142-dde312d0d6a2
 	github.com/tcnksm/go-httpstat v0.2.0
 	github.com/toqueteos/webbrowser v1.2.0
 	go4.org/mem v0.0.0-20201119185036-c04c5a6ff174
-	golang.org/x/crypto v0.0.0-20210317152858-513c2a44f670
-	golang.org/x/net v0.0.0-20210226172049-e18ecbb05110
+	golang.org/x/crypto v0.0.0-20210616213533-5ff15b29337e
+	golang.org/x/net v0.0.0-20210614182718-04defd469f4e
 	golang.org/x/sync v0.0.0-20210220032951-036812b2e83c
-	golang.org/x/sys v0.0.0-20210403161142-5e06dd20ab57
-	golang.org/x/term v0.0.0-20210317153231-de623e64d2a6
-	golang.org/x/time v0.0.0-20210220033141-f8bda1e9f3ba
-	golang.org/x/tools v0.0.0-20201211185031-d93e913c1a58
-	golang.zx2c4.com/wireguard/windows v0.1.2-0.20201113162609-9b85be97fdf8
-	gopkg.in/yaml.v2 v2.2.8 // indirect
-	honnef.co/go/tools v0.1.0
-	inet.af/netaddr v0.0.0-20210222205655-a1ec2b7b8c44
-	inet.af/netstack v0.0.0-20210317161235-a1bf4e56ef22
-	inet.af/peercred v0.0.0-20210302202138-56e694897155
+	golang.org/x/sys v0.0.0-20210616094352-59db8d763f22
+	golang.org/x/term v0.0.0-20210503060354-a79de5458b56
+	golang.org/x/time v0.0.0-20210611083556-38a9dc6acbc6
+	golang.org/x/tools v0.1.2
+	golang.zx2c4.com/wireguard v0.0.0-20210624150102-15b24b6179e0
+	golang.zx2c4.com/wireguard/windows v0.3.16
+	honnef.co/go/tools v0.1.4
+	inet.af/netaddr v0.0.0-20210721214506-ce7a8ad02cc1
+	inet.af/netstack v0.0.0-20210622165351-29b14ebc044e
+	inet.af/peercred v0.0.0-20210318190834-4259e17bb763
+	inet.af/wf v0.0.0-20210516214145-a5343001b756
 	rsc.io/goversion v1.2.0
 )
-
-replace github.com/github/certstore => github.com/cyolosecurity/certstore v0.0.0-20200922073901-ece7f1d353c2

health/health.go

@@ -11,6 +11,7 @@ import (
 	"fmt"
 	"sort"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/go-multierror/multierror"
@@ -36,6 +37,7 @@ var (
 	ipnState                string
 	ipnWantRunning          bool
 	anyInterfaceUp          = true // until told otherwise
+	udp4Unbound             bool
 )
 
 // Subsystem is the name of a subsystem whose health can be monitored.
@@ -46,7 +48,7 @@ const (
 	// the system, rather than one particular subsystem.
 	SysOverall = Subsystem("overall")
 
-	// SysRouter is the name the wgengine/router subsystem.
+	// SysRouter is the name of the wgengine/router subsystem.
 	SysRouter = Subsystem("router")
 
 	// SysDNS is the name of the net/dns subsystem.
@@ -213,9 +215,18 @@ func SetAnyInterfaceUp(up bool) {
 	selfCheckLocked()
 }
 
+// SetUDP4Unbound sets whether the udp4 bind failed completely.
+func SetUDP4Unbound(unbound bool) {
+	mu.Lock()
+	defer mu.Unlock()
+	udp4Unbound = unbound
+	selfCheckLocked()
+}
+
 func timerSelfCheck() {
 	mu.Lock()
 	defer mu.Unlock()
+	checkReceiveFuncs()
 	selfCheckLocked()
 	if timer != nil {
 		timer.Reset(time.Minute)
@@ -255,6 +266,9 @@ func overallErrorLocked() error {
 	if d := now.Sub(derpRegionLastFrame[rid]).Round(time.Second); d > tooIdle {
 		return fmt.Errorf("haven't heard from home DERP region %v in %v", rid, d)
 	}
+	if udp4Unbound {
+		return errors.New("no udp4 bind")
+	}
 
 	// TODO: use
 	_ = inMapPollSince
@@ -263,6 +277,11 @@ func overallErrorLocked() error {
 	_ = lastMapRequestHeard
 
 	var errs []error
+	for _, recv := range receiveFuncs {
+		if recv.missing {
+			errs = append(errs, fmt.Errorf("%s is not running", recv.name))
+		}
+	}
 	for sys, err := range sysErr {
 		if err == nil || sys == SysOverall {
 			continue
@@ -275,3 +294,58 @@ func overallErrorLocked() error {
 	})
 	return multierror.New(errs)
 }
+
+var (
+	ReceiveIPv4 = ReceiveFuncStats{name: "ReceiveIPv4"}
+	ReceiveIPv6 = ReceiveFuncStats{name: "ReceiveIPv6"}
+	ReceiveDERP = ReceiveFuncStats{name: "ReceiveDERP"}
+
+	receiveFuncs = []*ReceiveFuncStats{&ReceiveIPv4, &ReceiveIPv6, &ReceiveDERP}
+)
+
+// ReceiveFuncStats tracks the calls made to a wireguard-go receive func.
+type ReceiveFuncStats struct {
+	// name is the name of the receive func.
+	name string
+	// numCalls is the number of times the receive func has ever been called.
+	// It is required because it is possible for a receive func's wireguard-go goroutine
+	// to be active even though the receive func isn't.
+	// The wireguard-go goroutine alternates between calling the receive func and
+	// processing what the func returned.
+	numCalls uint64 // accessed atomically
+	// prevNumCalls is the value of numCalls last time the health check examined it.
+	prevNumCalls uint64
+	// inCall indicates whether the receive func is currently running.
+	inCall uint32 // bool, accessed atomically
+	// missing indicates whether the receive func is not running.
+	missing bool
+}
+
+func (s *ReceiveFuncStats) Enter() {
+	atomic.AddUint64(&s.numCalls, 1)
+	atomic.StoreUint32(&s.inCall, 1)
+}
+
+func (s *ReceiveFuncStats) Exit() {
+	atomic.StoreUint32(&s.inCall, 0)
+}
+
+func checkReceiveFuncs() {
+	for _, recv := range receiveFuncs {
+		recv.missing = false
+		prev := recv.prevNumCalls
+		numCalls := atomic.LoadUint64(&recv.numCalls)
+		recv.prevNumCalls = numCalls
+		if numCalls > prev {
+			// OK: the function has gotten called since last we checked
+			continue
+		}
+		if atomic.LoadUint32(&recv.inCall) == 1 {
+			// OK: the function is active, probably blocked due to inactivity
+			continue
+		}
+		// Not OK: The function is not active, and not accumulating new calls.
+		// It is probably MIA.
+		recv.missing = true
+	}
+}

hostinfo/hostinfo.go

@@ -0,0 +1,128 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package hostinfo answers questions about the host environment that Tailscale is
+// running on.
+//
+// TODO(bradfitz): move more of control/controlclient/hostinfo_* into this package.
+package hostinfo
+
+import (
+	"io"
+	"os"
+	"runtime"
+	"sync/atomic"
+
+	"go4.org/mem"
+	"tailscale.com/util/lineread"
+)
+
+// EnvType represents a known environment type.
+// The empty string, the default, means unknown.
+type EnvType string
+
+const (
+	KNative         = EnvType("kn")
+	AWSLambda       = EnvType("lm")
+	Heroku          = EnvType("hr")
+	AzureAppService = EnvType("az")
+	AWSFargate      = EnvType("fg")
+)
+
+var envType atomic.Value // of EnvType
+
+func GetEnvType() EnvType {
+	if e, ok := envType.Load().(EnvType); ok {
+		return e
+	}
+	e := getEnvType()
+	envType.Store(e)
+	return e
+}
+
+func getEnvType() EnvType {
+	if inKnative() {
+		return KNative
+	}
+	if inAWSLambda() {
+		return AWSLambda
+	}
+	if inHerokuDyno() {
+		return Heroku
+	}
+	if inAzureAppService() {
+		return AzureAppService
+	}
+	if inAWSFargate() {
+		return AWSFargate
+	}
+	return ""
+}
+
+// InContainer reports whether we're running in a container.
+func InContainer() bool {
+	if runtime.GOOS != "linux" {
+		return false
+	}
+	var ret bool
+	lineread.File("/proc/1/cgroup", func(line []byte) error {
+		if mem.Contains(mem.B(line), mem.S("/docker/")) ||
+			mem.Contains(mem.B(line), mem.S("/lxc/")) {
+			ret = true
+			return io.EOF // arbitrary non-nil error to stop loop
+		}
+		return nil
+	})
+	lineread.File("/proc/mounts", func(line []byte) error {
+		if mem.Contains(mem.B(line), mem.S("fuse.lxcfs")) {
+			ret = true
+			return io.EOF
+		}
+		return nil
+	})
+	return ret
+}
+
+func inKnative() bool {
+	// https://cloud.google.com/run/docs/reference/container-contract#env-vars
+	if os.Getenv("K_REVISION") != "" && os.Getenv("K_CONFIGURATION") != "" &&
+		os.Getenv("K_SERVICE") != "" && os.Getenv("PORT") != "" {
+		return true
+	}
+	return false
+}
+
+func inAWSLambda() bool {
+	// https://docs.aws.amazon.com/lambda/latest/dg/configuration-envvars.html
+	if os.Getenv("AWS_LAMBDA_FUNCTION_NAME") != "" &&
+		os.Getenv("AWS_LAMBDA_FUNCTION_VERSION") != "" &&
+		os.Getenv("AWS_LAMBDA_INITIALIZATION_TYPE") != "" &&
+		os.Getenv("AWS_LAMBDA_RUNTIME_API") != "" {
+		return true
+	}
+	return false
+}
+
+func inHerokuDyno() bool {
+	// https://devcenter.heroku.com/articles/dynos#local-environment-variables
+	if os.Getenv("PORT") != "" && os.Getenv("DYNO") != "" {
+		return true
+	}
+	return false
+}
+
+func inAzureAppService() bool {
+	if os.Getenv("APPSVC_RUN_ZIP") != "" && os.Getenv("WEBSITE_STACK") != "" &&
+		os.Getenv("WEBSITE_AUTH_AUTO_AAD") != "" {
+		return true
+	}
+	return false
+}
+
+func inAWSFargate() bool {
+	if os.Getenv("AWS_EXECUTION_ENV") == "AWS_ECS_FARGATE" {
+		return true
+	}
+	return false
+}

internal/deepprint/deepprint.go

@@ -1,103 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package deepprint walks a Go value recursively, in a predictable
-// order, without looping, and prints each value out to a given
-// Writer, which is assumed to be a hash.Hash, as this package doesn't
-// format things nicely.
-//
-// This is intended as a lighter version of go-spew, etc. We don't need its
-// features when our writer is just a hash.
-package deepprint
-
-import (
-	"crypto/sha256"
-	"fmt"
-	"io"
-	"reflect"
-)
-
-func Hash(v ...interface{}) string {
-	h := sha256.New()
-	Print(h, v)
-	return fmt.Sprintf("%x", h.Sum(nil))
-}
-
-// UpdateHash sets last to the hash of v and reports whether its value changed.
-func UpdateHash(last *string, v ...interface{}) (changed bool) {
-	sig := Hash(v)
-	if *last != sig {
-		*last = sig
-		return true
-	}
-	return false
-}
-
-func Print(w io.Writer, v ...interface{}) {
-	print(w, reflect.ValueOf(v), make(map[uintptr]bool))
-}
-
-func print(w io.Writer, v reflect.Value, visited map[uintptr]bool) {
-	if !v.IsValid() {
-		return
-	}
-	switch v.Kind() {
-	default:
-		panic(fmt.Sprintf("unhandled kind %v for type %v", v.Kind(), v.Type()))
-	case reflect.Ptr:
-		ptr := v.Pointer()
-		if visited[ptr] {
-			return
-		}
-		visited[ptr] = true
-		print(w, v.Elem(), visited)
-		return
-	case reflect.Struct:
-		fmt.Fprintf(w, "struct{\n")
-		t := v.Type()
-		for i, n := 0, v.NumField(); i < n; i++ {
-			sf := t.Field(i)
-			fmt.Fprintf(w, "%s: ", sf.Name)
-			print(w, v.Field(i), visited)
-			fmt.Fprintf(w, "\n")
-		}
-	case reflect.Slice, reflect.Array:
-		if v.Type().Elem().Kind() == reflect.Uint8 && v.CanInterface() {
-			fmt.Fprintf(w, "%q", v.Interface())
-			return
-		}
-		fmt.Fprintf(w, "[%d]{\n", v.Len())
-		for i, ln := 0, v.Len(); i < ln; i++ {
-			fmt.Fprintf(w, " [%d]: ", i)
-			print(w, v.Index(i), visited)
-			fmt.Fprintf(w, "\n")
-		}
-		fmt.Fprintf(w, "}\n")
-	case reflect.Interface:
-		print(w, v.Elem(), visited)
-	case reflect.Map:
-		sm := newSortedMap(v)
-		fmt.Fprintf(w, "map[%d]{\n", len(sm.Key))
-		for i, k := range sm.Key {
-			print(w, k, visited)
-			fmt.Fprintf(w, ": ")
-			print(w, sm.Value[i], visited)
-			fmt.Fprintf(w, "\n")
-		}
-		fmt.Fprintf(w, "}\n")
-
-	case reflect.String:
-		fmt.Fprintf(w, "%s", v.String())
-	case reflect.Bool:
-		fmt.Fprintf(w, "%v", v.Bool())
-	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
-		fmt.Fprintf(w, "%v", v.Int())
-	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
-		fmt.Fprintf(w, "%v", v.Uint())
-	case reflect.Float32, reflect.Float64:
-		fmt.Fprintf(w, "%v", v.Float())
-	case reflect.Complex64, reflect.Complex128:
-		fmt.Fprintf(w, "%v", v.Complex())
-	}
-}

internal/deepprint/deepprint_test.go

@@ -1,64 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package deepprint
-
-import (
-	"bytes"
-	"testing"
-
-	"inet.af/netaddr"
-	"tailscale.com/wgengine/router"
-	"tailscale.com/wgengine/wgcfg"
-)
-
-func TestDeepPrint(t *testing.T) {
-	// v contains the types of values we care about for our current callers.
-	// Mostly we're just testing that we don't panic on handled types.
-	v := getVal()
-
-	var buf bytes.Buffer
-	Print(&buf, v)
-	t.Logf("Got: %s", buf.Bytes())
-
-	hash1 := Hash(v)
-	t.Logf("hash: %v", hash1)
-	for i := 0; i < 20; i++ {
-		hash2 := Hash(getVal())
-		if hash1 != hash2 {
-			t.Error("second hash didn't match")
-		}
-	}
-}
-
-func getVal() []interface{} {
-	return []interface{}{
-		&wgcfg.Config{
-			Name:      "foo",
-			Addresses: []netaddr.IPPrefix{{Bits: 5, IP: netaddr.IPFrom16([16]byte{3: 3})}},
-			Peers: []wgcfg.Peer{
-				{
-					Endpoints: "foo:5",
-				},
-			},
-		},
-		&router.Config{
-			Routes: []netaddr.IPPrefix{
-				netaddr.MustParseIPPrefix("1.2.3.0/24"),
-				netaddr.MustParseIPPrefix("1234::/64"),
-			},
-		},
-		map[string]string{
-			"key1": "val1",
-			"key2": "val2",
-			"key3": "val3",
-			"key4": "val4",
-			"key5": "val5",
-			"key6": "val6",
-			"key7": "val7",
-			"key8": "val8",
-			"key9": "val9",
-		},
-	}
-}

internal/deepprint/fmtsort.go

@@ -1,224 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// and
-
-// Copyright 2018 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// This is a slightly modified fork of Go's src/internal/fmtsort/sort.go
-
-package deepprint
-
-import (
-	"reflect"
-	"sort"
-)
-
-// Note: Throughout this package we avoid calling reflect.Value.Interface as
-// it is not always legal to do so and it's easier to avoid the issue than to face it.
-
-// sortedMap represents a map's keys and values. The keys and values are
-// aligned in index order: Value[i] is the value in the map corresponding to Key[i].
-type sortedMap struct {
-	Key   []reflect.Value
-	Value []reflect.Value
-}
-
-func (o *sortedMap) Len() int           { return len(o.Key) }
-func (o *sortedMap) Less(i, j int) bool { return compare(o.Key[i], o.Key[j]) < 0 }
-func (o *sortedMap) Swap(i, j int) {
-	o.Key[i], o.Key[j] = o.Key[j], o.Key[i]
-	o.Value[i], o.Value[j] = o.Value[j], o.Value[i]
-}
-
-// Sort accepts a map and returns a sortedMap that has the same keys and
-// values but in a stable sorted order according to the keys, modulo issues
-// raised by unorderable key values such as NaNs.
-//
-// The ordering rules are more general than with Go's < operator:
-//
-//  - when applicable, nil compares low
-//  - ints, floats, and strings order by <
-//  - NaN compares less than non-NaN floats
-//  - bool compares false before true
-//  - complex compares real, then imag
-//  - pointers compare by machine address
-//  - channel values compare by machine address
-//  - structs compare each field in turn
-//  - arrays compare each element in turn.
-//    Otherwise identical arrays compare by length.
-//  - interface values compare first by reflect.Type describing the concrete type
-//    and then by concrete value as described in the previous rules.
-//
-func newSortedMap(mapValue reflect.Value) *sortedMap {
-	if mapValue.Type().Kind() != reflect.Map {
-		return nil
-	}
-	// Note: this code is arranged to not panic even in the presence
-	// of a concurrent map update. The runtime is responsible for
-	// yelling loudly if that happens. See issue 33275.
-	n := mapValue.Len()
-	key := make([]reflect.Value, 0, n)
-	value := make([]reflect.Value, 0, n)
-	iter := mapValue.MapRange()
-	for iter.Next() {
-		key = append(key, iter.Key())
-		value = append(value, iter.Value())
-	}
-	sorted := &sortedMap{
-		Key:   key,
-		Value: value,
-	}
-	sort.Stable(sorted)
-	return sorted
-}
-
-// compare compares two values of the same type. It returns -1, 0, 1
-// according to whether a > b (1), a == b (0), or a < b (-1).
-// If the types differ, it returns -1.
-// See the comment on Sort for the comparison rules.
-func compare(aVal, bVal reflect.Value) int {
-	aType, bType := aVal.Type(), bVal.Type()
-	if aType != bType {
-		return -1 // No good answer possible, but don't return 0: they're not equal.
-	}
-	switch aVal.Kind() {
-	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
-		a, b := aVal.Int(), bVal.Int()
-		switch {
-		case a < b:
-			return -1
-		case a > b:
-			return 1
-		default:
-			return 0
-		}
-	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
-		a, b := aVal.Uint(), bVal.Uint()
-		switch {
-		case a < b:
-			return -1
-		case a > b:
-			return 1
-		default:
-			return 0
-		}
-	case reflect.String:
-		a, b := aVal.String(), bVal.String()
-		switch {
-		case a < b:
-			return -1
-		case a > b:
-			return 1
-		default:
-			return 0
-		}
-	case reflect.Float32, reflect.Float64:
-		return floatCompare(aVal.Float(), bVal.Float())
-	case reflect.Complex64, reflect.Complex128:
-		a, b := aVal.Complex(), bVal.Complex()
-		if c := floatCompare(real(a), real(b)); c != 0 {
-			return c
-		}
-		return floatCompare(imag(a), imag(b))
-	case reflect.Bool:
-		a, b := aVal.Bool(), bVal.Bool()
-		switch {
-		case a == b:
-			return 0
-		case a:
-			return 1
-		default:
-			return -1
-		}
-	case reflect.Ptr:
-		a, b := aVal.Pointer(), bVal.Pointer()
-		switch {
-		case a < b:
-			return -1
-		case a > b:
-			return 1
-		default:
-			return 0
-		}
-	case reflect.Chan:
-		if c, ok := nilCompare(aVal, bVal); ok {
-			return c
-		}
-		ap, bp := aVal.Pointer(), bVal.Pointer()
-		switch {
-		case ap < bp:
-			return -1
-		case ap > bp:
-			return 1
-		default:
-			return 0
-		}
-	case reflect.Struct:
-		for i := 0; i < aVal.NumField(); i++ {
-			if c := compare(aVal.Field(i), bVal.Field(i)); c != 0 {
-				return c
-			}
-		}
-		return 0
-	case reflect.Array:
-		for i := 0; i < aVal.Len(); i++ {
-			if c := compare(aVal.Index(i), bVal.Index(i)); c != 0 {
-				return c
-			}
-		}
-		return 0
-	case reflect.Interface:
-		if c, ok := nilCompare(aVal, bVal); ok {
-			return c
-		}
-		c := compare(reflect.ValueOf(aVal.Elem().Type()), reflect.ValueOf(bVal.Elem().Type()))
-		if c != 0 {
-			return c
-		}
-		return compare(aVal.Elem(), bVal.Elem())
-	default:
-		// Certain types cannot appear as keys (maps, funcs, slices), but be explicit.
-		panic("bad type in compare: " + aType.String())
-	}
-}
-
-// nilCompare checks whether either value is nil. If not, the boolean is false.
-// If either value is nil, the boolean is true and the integer is the comparison
-// value. The comparison is defined to be 0 if both are nil, otherwise the one
-// nil value compares low. Both arguments must represent a chan, func,
-// interface, map, pointer, or slice.
-func nilCompare(aVal, bVal reflect.Value) (int, bool) {
-	if aVal.IsNil() {
-		if bVal.IsNil() {
-			return 0, true
-		}
-		return -1, true
-	}
-	if bVal.IsNil() {
-		return 1, true
-	}
-	return 0, false
-}
-
-// floatCompare compares two floating-point values. NaNs compare low.
-func floatCompare(a, b float64) int {
-	switch {
-	case isNaN(a):
-		return -1 // No good answer if b is a NaN so don't bother checking.
-	case isNaN(b):
-		return 1
-	case a < b:
-		return -1
-	case a > b:
-		return 1
-	}
-	return 0
-}
-
-func isNaN(a float64) bool {
-	return a != a
-}

ipn/backend.go

@@ -5,7 +5,8 @@
 package ipn
 
 import (
-	"net/http"
+	"fmt"
+	"strings"
 	"time"
 
 	"tailscale.com/ipn/ipnstate"
@@ -56,20 +57,34 @@ type EngineStatus struct {
 // that they have not changed.
 // They are JSON-encoded on the wire, despite the lack of struct tags.
 type Notify struct {
-	_             structs.Incomparable
-	Version       string             // version number of IPN backend
-	ErrMessage    *string            // critical error message, if any; for InUseOtherUser, the details
-	LoginFinished *empty.Message     // event: non-nil when login process succeeded
-	State         *State             // current IPN state has changed
-	Prefs         *Prefs             // preferences were changed
-	NetMap        *netmap.NetworkMap // new netmap received
-	Engine        *EngineStatus      // wireguard engine stats
-	BrowseToURL   *string            // UI should open a browser right now
-	BackendLogID  *string            // public logtail id used by backend
-	PingResult    *ipnstate.PingResult
+	_       structs.Incomparable
+	Version string // version number of IPN backend
 
+	// ErrMessage, if non-nil, contains a critical error message.
+	// For State InUseOtherUser, ErrMessage is not critical and just contains the details.
+	ErrMessage *string
+
+	LoginFinished *empty.Message       // non-nil when/if the login process succeeded
+	State         *State               // if non-nil, the new or current IPN state
+	Prefs         *Prefs               // if non-nil, the new or current preferences
+	NetMap        *netmap.NetworkMap   // if non-nil, the new or current netmap
+	Engine        *EngineStatus        // if non-nil, the new or urrent wireguard stats
+	BrowseToURL   *string              // if non-nil, UI should open a browser right now
+	BackendLogID  *string              // if non-nil, the public logtail ID used by backend
+	PingResult    *ipnstate.PingResult // if non-nil, a ping response arrived
+
+	// FilesWaiting if non-nil means that files are buffered in
+	// the Tailscale daemon and ready for local transfer to the
+	// user's preferred storage location.
 	FilesWaiting *empty.Message `json:",omitempty"`
 
+	// IncomingFiles, if non-nil, specifies which files are in the
+	// process of being received. A nil IncomingFiles means this
+	// Notify should not update the state of file transfers. A non-nil
+	// but empty IncomingFiles means that no files are in the middle
+	// of being transferred.
+	IncomingFiles []PartialFile `json:",omitempty"`
+
 	// LocalTCPPort, if non-nil, informs the UI frontend which
 	// (non-zero) localhost TCP port it's listening on.
 	// This is currently only used by Tailscale when run in the
@@ -79,6 +94,67 @@ type Notify struct {
 	// type is mirrored in xcode/Shared/IPN.swift
 }
 
+func (n Notify) String() string {
+	var sb strings.Builder
+	sb.WriteString("Notify{")
+	if n.ErrMessage != nil {
+		fmt.Fprintf(&sb, "err=%q ", *n.ErrMessage)
+	}
+	if n.LoginFinished != nil {
+		sb.WriteString("LoginFinished ")
+	}
+	if n.State != nil {
+		fmt.Fprintf(&sb, "state=%v ", *n.State)
+	}
+	if n.Prefs != nil {
+		fmt.Fprintf(&sb, "%v ", n.Prefs.Pretty())
+	}
+	if n.NetMap != nil {
+		sb.WriteString("NetMap{...} ")
+	}
+	if n.Engine != nil {
+		fmt.Fprintf(&sb, "wg=%v ", *n.Engine)
+	}
+	if n.BrowseToURL != nil {
+		sb.WriteString("URL=<...> ")
+	}
+	if n.BackendLogID != nil {
+		sb.WriteString("BackendLogID ")
+	}
+	if n.PingResult != nil {
+		fmt.Fprintf(&sb, "ping=%v ", *n.PingResult)
+	}
+	if n.FilesWaiting != nil {
+		sb.WriteString("FilesWaiting ")
+	}
+	if len(n.IncomingFiles) != 0 {
+		sb.WriteString("IncomingFiles ")
+	}
+	if n.LocalTCPPort != nil {
+		fmt.Fprintf(&sb, "tcpport=%v ", n.LocalTCPPort)
+	}
+	s := sb.String()
+	return s[0:len(s)-1] + "}"
+}
+
+// PartialFile represents an in-progress file transfer.
+type PartialFile struct {
+	Name         string    // e.g. "foo.jpg"
+	Started      time.Time // time transfer started
+	DeclaredSize int64     // or -1 if unknown
+	Received     int64     // bytes copied thus far
+
+	// PartialPath is set non-empty in "direct" file mode to the
+	// in-progress '*.partial' file's path when the peerapi isn't
+	// being used; see LocalBackend.SetDirectFileRoot.
+	PartialPath string `json:",omitempty"`
+
+	// Done is set in "direct" mode when the partial file has been
+	// closed and is ready for the caller to rename away the
+	// ".partial" suffix.
+	Done bool `json:",omitempty"`
+}
+
 // StateKey is an opaque identifier for a set of LocalBackend state
 // (preferences, private keys, etc.).
 //
@@ -109,24 +185,33 @@ type Options struct {
 	//    state and use/update that.
 	//  - StateKey!="" && Prefs!=nil: like the previous case, but do
 	//    an initial overwrite of backend state with Prefs.
+	//
+	// NOTE(apenwarr): The above means that this Prefs field does not do
+	// what you probably think it does. It will overwrite your encryption
+	// keys. Do not use unless you know what you're doing.
 	StateKey StateKey
 	Prefs    *Prefs
+	// UpdatePrefs, if provided, overrides Options.Prefs *and* the Prefs
+	// already stored in the backend state, *except* for the Persist
+	// Persist member. If you just want to provide prefs, this is
+	// probably what you want.
+	//
+	// UpdatePrefs.Persist is always ignored. Prefs.Persist will still
+	// be used even if UpdatePrefs is provided. Other than Persist,
+	// UpdatePrefs takes precedence over Prefs.
+	//
+	// This is intended as a purely temporary workaround for the
+	// currently unexpected behaviour of Options.Prefs.
+	//
+	// TODO(apenwarr): Remove this, or rename Prefs to something else
+	//   and rename this to Prefs. Or, move Prefs.Persist elsewhere
+	//   entirely (as it always should have been), and then we wouldn't
+	//   need two separate fields at all. Or, move the fancy state
+	//   migration stuff out of Start().
+	UpdatePrefs *Prefs
 	// AuthKey is an optional node auth key used to authorize a
 	// new node key without user interaction.
 	AuthKey string
-	// LegacyConfigPath optionally specifies the old-style relaynode
-	// relay.conf location. If both LegacyConfigPath and StateKey are
-	// specified and the requested state doesn't exist in the backend
-	// store, the backend migrates the config from LegacyConfigPath.
-	//
-	// TODO(danderson): remove some time after the transition to
-	// tailscaled is done.
-	LegacyConfigPath string
-	// Notify is called when backend events happen.
-	Notify func(Notify) `json:"-"`
-	// HTTPTestClient is an optional HTTP client to pass to controlclient
-	// (for tests only).
-	HTTPTestClient *http.Client
 }
 
 // Backend is the interface between Tailscale frontends
@@ -135,6 +220,9 @@ type Options struct {
 // (It has nothing to do with the interface between the backends
 // and the cloud control plane.)
 type Backend interface {
+	// SetNotifyCallback sets the callback to be called on updates
+	// from the backend to the client.
+	SetNotifyCallback(func(Notify))
 	// Start starts or restarts the backend, typically when a
 	// frontend client connects.
 	Start(Options) error
@@ -151,8 +239,6 @@ type Backend interface {
 	// WantRunning. This may cause the wireguard engine to
 	// reconfigure or stop.
 	SetPrefs(*Prefs)
-	// EditPrefs is like SetPrefs but only sets the specified fields.
-	EditPrefs(*MaskedPrefs)
 	// RequestEngineStatus polls for an update from the wireguard
 	// engine. Only needed if you want to display byte
 	// counts. Connection events are emitted automatically without

ipn/fake_test.go

@@ -5,7 +5,6 @@
 package ipn
 
 import (
-	"log"
 	"time"
 
 	"tailscale.com/ipn/ipnstate"
@@ -20,19 +19,29 @@ type FakeBackend struct {
 }
 
 func (b *FakeBackend) Start(opts Options) error {
-	b.serverURL = opts.Prefs.ControlURL
-	if opts.Notify == nil {
-		log.Fatalf("FakeBackend.Start: opts.Notify is nil\n")
+	b.serverURL = opts.Prefs.ControlURLOrDefault()
+	if b.notify == nil {
+		panic("FakeBackend.Start: SetNotifyCallback not called")
 	}
-	b.notify = opts.Notify
-	b.notify(Notify{Prefs: opts.Prefs})
 	nl := NeedsLogin
-	b.notify(Notify{State: &nl})
+	if b.notify != nil {
+		b.notify(Notify{Prefs: opts.Prefs})
+		b.notify(Notify{State: &nl})
+	}
 	return nil
 }
 
+func (b *FakeBackend) SetNotifyCallback(notify func(Notify)) {
+	if notify == nil {
+		panic("FakeBackend.SetNotifyCallback: notify is nil")
+	}
+	b.notify = notify
+}
+
 func (b *FakeBackend) newState(s State) {
-	b.notify(Notify{State: &s})
+	if b.notify != nil {
+		b.notify(Notify{State: &s})
+	}
 	if s == Running {
 		b.live = true
 	} else {
@@ -42,7 +51,9 @@ func (b *FakeBackend) newState(s State) {
 
 func (b *FakeBackend) StartLoginInteractive() {
 	u := b.serverURL + "/this/is/fake"
-	b.notify(Notify{BrowseToURL: &u})
+	if b.notify != nil {
+		b.notify(Notify{BrowseToURL: &u})
+	}
 	b.login()
 }
 
@@ -54,10 +65,14 @@ func (b *FakeBackend) login() {
 	b.newState(NeedsMachineAuth)
 	b.newState(Stopped)
 	// TODO(apenwarr): Fill in a more interesting netmap here.
-	b.notify(Notify{NetMap: &netmap.NetworkMap{}})
+	if b.notify != nil {
+		b.notify(Notify{NetMap: &netmap.NetworkMap{}})
+	}
 	b.newState(Starting)
 	// TODO(apenwarr): Fill in a more interesting status.
-	b.notify(Notify{Engine: &EngineStatus{}})
+	if b.notify != nil {
+		b.notify(Notify{Engine: &EngineStatus{}})
+	}
 	b.newState(Running)
 }
 
@@ -70,7 +85,9 @@ func (b *FakeBackend) SetPrefs(new *Prefs) {
 		panic("FakeBackend.SetPrefs got nil prefs")
 	}
 
-	b.notify(Notify{Prefs: new.Clone()})
+	if b.notify != nil {
+		b.notify(Notify{Prefs: new.Clone()})
+	}
 	if new.WantRunning && !b.live {
 		b.newState(Starting)
 		b.newState(Running)
@@ -79,21 +96,20 @@ func (b *FakeBackend) SetPrefs(new *Prefs) {
 	}
 }
 
-func (b *FakeBackend) EditPrefs(mp *MaskedPrefs) {
-	// This fake implementation only cares about this one pref.
-	if mp.WantRunningSet {
-		b.SetPrefs(&mp.Prefs)
+func (b *FakeBackend) RequestEngineStatus() {
+	if b.notify != nil {
+		b.notify(Notify{Engine: &EngineStatus{}})
 	}
 }
 
-func (b *FakeBackend) RequestEngineStatus() {
-	b.notify(Notify{Engine: &EngineStatus{}})
-}
-
 func (b *FakeBackend) FakeExpireAfter(x time.Duration) {
-	b.notify(Notify{NetMap: &netmap.NetworkMap{}})
+	if b.notify != nil {
+		b.notify(Notify{NetMap: &netmap.NetworkMap{}})
+	}
 }
 
 func (b *FakeBackend) Ping(ip string, useTSMP bool) {
-	b.notify(Notify{PingResult: &ipnstate.PingResult{}})
+	if b.notify != nil {
+		b.notify(Notify{PingResult: &ipnstate.PingResult{}})
+	}
 }

ipn/handle.go

@@ -15,25 +15,26 @@ import (
 )
 
 type Handle struct {
-	frontendLogID string
-	b             Backend
-	xnotify       func(Notify)
-	logf          logger.Logf
+	b    Backend
+	logf logger.Logf
 
 	// Mutex protects everything below
 	mu                sync.Mutex
+	xnotify           func(Notify)
+	frontendLogID     string
 	netmapCache       *netmap.NetworkMap
 	engineStatusCache EngineStatus
 	stateCache        State
 	prefsCache        *Prefs
 }
 
-func NewHandle(b Backend, logf logger.Logf, opts Options) (*Handle, error) {
+func NewHandle(b Backend, logf logger.Logf, notify func(Notify), opts Options) (*Handle, error) {
 	h := &Handle{
 		b:    b,
 		logf: logf,
 	}
 
+	h.SetNotifyCallback(notify)
 	err := h.Start(opts)
 	if err != nil {
 		return nil, err
@@ -42,18 +43,25 @@ func NewHandle(b Backend, logf logger.Logf, opts Options) (*Handle, error) {
 	return h, nil
 }
 
+func (h *Handle) SetNotifyCallback(notify func(Notify)) {
+	h.mu.Lock()
+	h.xnotify = notify
+	h.mu.Unlock()
+
+	h.b.SetNotifyCallback(h.notify)
+}
+
 func (h *Handle) Start(opts Options) error {
+	h.mu.Lock()
 	h.frontendLogID = opts.FrontendLogID
-	h.xnotify = opts.Notify
 	h.netmapCache = nil
 	h.engineStatusCache = EngineStatus{}
 	h.stateCache = NoState
 	if opts.Prefs != nil {
 		h.prefsCache = opts.Prefs.Clone()
 	}
-	xopts := opts
-	xopts.Notify = h.notify
-	return h.b.Start(xopts)
+	h.mu.Unlock()
+	return h.b.Start(opts)
 }
 
 func (h *Handle) Reset() {
@@ -148,7 +156,7 @@ func (h *Handle) Expiry() time.Time {
 }
 
 func (h *Handle) AdminPageURL() string {
-	return h.prefsCache.ControlURL + "/admin/machines"
+	return h.prefsCache.AdminPageURL()
 }
 
 func (h *Handle) StartLoginInteractive() {

ipn/ipnlocal/local_test.go

@@ -5,14 +5,20 @@
 package ipnlocal
 
 import (
+	"fmt"
+	"net/http"
 	"reflect"
 	"testing"
+	"time"
 
 	"inet.af/netaddr"
+	"tailscale.com/ipn"
 	"tailscale.com/net/interfaces"
 	"tailscale.com/net/tsaddr"
 	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
 	"tailscale.com/types/netmap"
+	"tailscale.com/wgengine"
 	"tailscale.com/wgengine/wgcfg"
 )
 
@@ -165,7 +171,7 @@ func TestShrinkDefaultRoute(t *testing.T) {
 			out: []string{
 				"fe80::1",
 				"ff00::1",
-				tsaddr.TailscaleULARange().IP.String(),
+				tsaddr.TailscaleULARange().IP().String(),
 			},
 			localIPFn: func(ip netaddr.IP) bool { return !inRemove(ip) && ip.Is6() },
 		},
@@ -291,3 +297,199 @@ func TestPeerRoutes(t *testing.T) {
 	}
 
 }
+
+func TestPeerAPIBase(t *testing.T) {
+	tests := []struct {
+		name string
+		nm   *netmap.NetworkMap
+		peer *tailcfg.Node
+		want string
+	}{
+		{
+			name: "nil_netmap",
+			peer: new(tailcfg.Node),
+			want: "",
+		},
+		{
+			name: "nil_peer",
+			nm:   new(netmap.NetworkMap),
+			want: "",
+		},
+		{
+			name: "self_only_4_them_both",
+			nm: &netmap.NetworkMap{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.1/32"),
+				},
+			},
+			peer: &tailcfg.Node{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.2/32"),
+					netaddr.MustParseIPPrefix("fe70::2/128"),
+				},
+				Hostinfo: tailcfg.Hostinfo{
+					Services: []tailcfg.Service{
+						{Proto: "peerapi4", Port: 444},
+						{Proto: "peerapi6", Port: 666},
+					},
+				},
+			},
+			want: "http://100.64.1.2:444",
+		},
+		{
+			name: "self_only_6_them_both",
+			nm: &netmap.NetworkMap{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("fe70::1/128"),
+				},
+			},
+			peer: &tailcfg.Node{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.2/32"),
+					netaddr.MustParseIPPrefix("fe70::2/128"),
+				},
+				Hostinfo: tailcfg.Hostinfo{
+					Services: []tailcfg.Service{
+						{Proto: "peerapi4", Port: 444},
+						{Proto: "peerapi6", Port: 666},
+					},
+				},
+			},
+			want: "http://[fe70::2]:666",
+		},
+		{
+			name: "self_both_them_only_4",
+			nm: &netmap.NetworkMap{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.1/32"),
+					netaddr.MustParseIPPrefix("fe70::1/128"),
+				},
+			},
+			peer: &tailcfg.Node{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.2/32"),
+					netaddr.MustParseIPPrefix("fe70::2/128"),
+				},
+				Hostinfo: tailcfg.Hostinfo{
+					Services: []tailcfg.Service{
+						{Proto: "peerapi4", Port: 444},
+					},
+				},
+			},
+			want: "http://100.64.1.2:444",
+		},
+		{
+			name: "self_both_them_only_6",
+			nm: &netmap.NetworkMap{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.1/32"),
+					netaddr.MustParseIPPrefix("fe70::1/128"),
+				},
+			},
+			peer: &tailcfg.Node{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.2/32"),
+					netaddr.MustParseIPPrefix("fe70::2/128"),
+				},
+				Hostinfo: tailcfg.Hostinfo{
+					Services: []tailcfg.Service{
+						{Proto: "peerapi6", Port: 666},
+					},
+				},
+			},
+			want: "http://[fe70::2]:666",
+		},
+		{
+			name: "self_both_them_no_peerapi_service",
+			nm: &netmap.NetworkMap{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.1/32"),
+					netaddr.MustParseIPPrefix("fe70::1/128"),
+				},
+			},
+			peer: &tailcfg.Node{
+				Addresses: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("100.64.1.2/32"),
+					netaddr.MustParseIPPrefix("fe70::2/128"),
+				},
+			},
+			want: "",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := peerAPIBase(tt.nm, tt.peer)
+			if got != tt.want {
+				t.Errorf("got %q; want %q", got, tt.want)
+			}
+		})
+	}
+}
+
+type panicOnUseTransport struct{}
+
+func (panicOnUseTransport) RoundTrip(*http.Request) (*http.Response, error) {
+	panic("unexpected HTTP request")
+}
+
+// Issue 1573: don't generate a machine key if we don't want to be running.
+func TestLazyMachineKeyGeneration(t *testing.T) {
+	defer func(old bool) { panicOnMachineKeyGeneration = old }(panicOnMachineKeyGeneration)
+	panicOnMachineKeyGeneration = true
+
+	var logf logger.Logf = logger.Discard
+	store := new(ipn.MemoryStore)
+	eng, err := wgengine.NewFakeUserspaceEngine(logf, 0)
+	if err != nil {
+		t.Fatalf("NewFakeUserspaceEngine: %v", err)
+	}
+	lb, err := NewLocalBackend(logf, "logid", store, eng)
+	if err != nil {
+		t.Fatalf("NewLocalBackend: %v", err)
+	}
+
+	lb.SetHTTPTestClient(&http.Client{
+		Transport: panicOnUseTransport{}, // validate we don't send HTTP requests
+	})
+
+	if err := lb.Start(ipn.Options{
+		StateKey: ipn.GlobalDaemonStateKey,
+	}); err != nil {
+		t.Fatalf("Start: %v", err)
+	}
+
+	// Give the controlclient package goroutines (if they're
+	// accidentally started) extra time to schedule and run (and thus
+	// hit panicOnUseTransport).
+	time.Sleep(500 * time.Millisecond)
+}
+
+func TestFileTargets(t *testing.T) {
+	b := new(LocalBackend)
+	_, err := b.FileTargets()
+	if got, want := fmt.Sprint(err), "not connected"; got != want {
+		t.Errorf("before connect: got %q; want %q", got, want)
+	}
+
+	b.netMap = new(netmap.NetworkMap)
+	_, err = b.FileTargets()
+	if got, want := fmt.Sprint(err), "not connected"; got != want {
+		t.Errorf("non-running netmap: got %q; want %q", got, want)
+	}
+
+	b.state = ipn.Running
+	_, err = b.FileTargets()
+	if got, want := fmt.Sprint(err), "file sharing not enabled by Tailscale admin"; got != want {
+		t.Errorf("without cap: got %q; want %q", got, want)
+	}
+
+	b.capFileSharing = true
+	got, err := b.FileTargets()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(got) != 0 {
+		t.Fatalf("unexpected %d peers", len(got))
+	}
+	// (other cases handled by TestPeerAPIBase above)
+}

ipn/ipnlocal/loglines_test.go

@@ -15,6 +15,7 @@ import (
 	"tailscale.com/tailcfg"
 	"tailscale.com/tstest"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
 	"tailscale.com/types/persist"
 	"tailscale.com/wgengine"
 )
@@ -30,6 +31,12 @@ func TestLocalLogLines(t *testing.T) {
 	})
 	defer logListen.Close()
 
+	// Put a rate-limiter with a burst of 0 between the components below.
+	// This instructs the rate-limiter to eliminate all logging that
+	// isn't explicitly exempt from rate-limiting.
+	// This lets the logListen tracker verify that the rate-limiter allows these key lines.
+	logf := logger.RateLimitedFnWithClock(logListen.Logf, 5*time.Second, 0, 10, time.Now)
+
 	logid := func(hex byte) logtail.PublicID {
 		var ret logtail.PublicID
 		for i := 0; i < len(ret); i++ {
@@ -41,12 +48,12 @@ func TestLocalLogLines(t *testing.T) {
 
 	// set up a LocalBackend, super bare bones. No functional data.
 	store := &ipn.MemoryStore{}
-	e, err := wgengine.NewFakeUserspaceEngine(logListen.Logf, 0)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, 0)
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	lb, err := NewLocalBackend(logListen.Logf, idA.String(), store, e)
+	lb, err := NewLocalBackend(logf, idA.String(), store, e)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -61,6 +68,7 @@ func TestLocalLogLines(t *testing.T) {
 	testWantRemain := func(wantRemain ...string) func(t *testing.T) {
 		return func(t *testing.T) {
 			if remain := logListen.Check(); !reflect.DeepEqual(remain, wantRemain) {
+				t.Helper()
 				t.Errorf("remain %q, want %q", remain, wantRemain)
 			}
 		}
@@ -75,18 +83,30 @@ func TestLocalLogLines(t *testing.T) {
 	t.Run("after_prefs", testWantRemain("[v1] peer keys: %s", "[v1] v%v peers: %v"))
 
 	// log peers, peer keys
-	status := &wgengine.Status{
+	lb.mu.Lock()
+	lb.parseWgStatusLocked(&wgengine.Status{
 		Peers: []ipnstate.PeerStatusLite{{
 			TxBytes:       10,
 			RxBytes:       10,
 			LastHandshake: time.Now(),
 			NodeKey:       tailcfg.NodeKey(key.NewPrivate()),
 		}},
-		LocalAddrs: []string{"idk an address"},
-	}
-	lb.mu.Lock()
-	lb.parseWgStatusLocked(status)
+	})
 	lb.mu.Unlock()
 
 	t.Run("after_peers", testWantRemain())
+
+	// Log it again with different stats to ensure it's not dup-suppressed.
+	logListen.Reset()
+	lb.mu.Lock()
+	lb.parseWgStatusLocked(&wgengine.Status{
+		Peers: []ipnstate.PeerStatusLite{{
+			TxBytes:       11,
+			RxBytes:       12,
+			LastHandshake: time.Now(),
+			NodeKey:       tailcfg.NodeKey(key.NewPrivate()),
+		}},
+	})
+	lb.mu.Unlock()
+	t.Run("after_second_peer_status", testWantRemain("SetPrefs: %v"))
 }

ipn/ipnlocal/peerapi.go

@@ -11,6 +11,7 @@ import (
 	"hash/crc32"
 	"html"
 	"io"
+	"io/fs"
 	"net"
 	"net/http"
 	"net/url"
@@ -18,11 +19,18 @@ import (
 	"path"
 	"path/filepath"
 	"runtime"
+	"sort"
 	"strconv"
 	"strings"
+	"sync"
+	"time"
+	"unicode"
+	"unicode/utf8"
 
 	"inet.af/netaddr"
+	"tailscale.com/client/tailscale/apitype"
 	"tailscale.com/ipn"
+	"tailscale.com/logtail/backoff"
 	"tailscale.com/net/interfaces"
 	"tailscale.com/syncs"
 	"tailscale.com/tailcfg"
@@ -37,25 +45,77 @@ type peerAPIServer struct {
 	tunName    string
 	selfNode   *tailcfg.Node
 	knownEmpty syncs.AtomicBool
+
+	// directFileMode is whether we're writing files directly to a
+	// download directory (as *.partial files), rather than making
+	// the frontend retrieve it over localapi HTTP and write it
+	// somewhere itself. This is used on GUI macOS version.
+	// In directFileMode, the peerapi doesn't do the final rename
+	// from "foo.jpg.partial" to "foo.jpg".
+	directFileMode bool
 }
 
-const partialSuffix = ".tspartial"
+const (
+	// partialSuffix is the suffix appened to files while they're
+	// still in the process of being transferred.
+	partialSuffix = ".partial"
+
+	// deletedSuffix is the suffix for a deleted marker file
+	// that's placed next to a file (without the suffix) that we
+	// tried to delete, but Windows wouldn't let us. These are
+	// only written on Windows (and in tests), but they're not
+	// permitted to be uploaded directly on any platform, like
+	// partial files.
+	deletedSuffix = ".deleted"
+)
+
+func validFilenameRune(r rune) bool {
+	switch r {
+	case '/':
+		return false
+	case '\\', ':', '*', '"', '<', '>', '|':
+		// Invalid stuff on Windows, but we reject them everywhere
+		// for now.
+		// TODO(bradfitz): figure out a better plan. We initially just
+		// wrote things to disk URL path-escaped, but that's gross
+		// when debugging, and just moves the problem to callers.
+		// So now we put the UTF-8 filenames on disk directly as
+		// sent.
+		return false
+	}
+	return unicode.IsPrint(r)
+}
 
 func (s *peerAPIServer) diskPath(baseName string) (fullPath string, ok bool) {
+	if !utf8.ValidString(baseName) {
+		return "", false
+	}
+	if strings.TrimSpace(baseName) != baseName {
+		return "", false
+	}
+	if len(baseName) > 255 {
+		return "", false
+	}
+	// TODO: validate unicode normalization form too? Varies by platform.
 	clean := path.Clean(baseName)
 	if clean != baseName ||
-		clean == "." ||
-		strings.ContainsAny(clean, `/\`) ||
+		clean == "." || clean == ".." ||
+		strings.HasSuffix(clean, deletedSuffix) ||
 		strings.HasSuffix(clean, partialSuffix) {
 		return "", false
 	}
-	return filepath.Join(s.rootDir, strings.ReplaceAll(url.PathEscape(baseName), ":", "%3a")), true
+	for _, r := range baseName {
+		if !validFilenameRune(r) {
+			return "", false
+		}
+	}
+	return filepath.Join(s.rootDir, baseName), true
 }
 
 // hasFilesWaiting reports whether any files are buffered in the
 // tailscaled daemon storage.
 func (s *peerAPIServer) hasFilesWaiting() bool {
-	if s.rootDir == "" {
+	if s.rootDir == "" || s.directFileMode {
 		return false
 	}
 	if s.knownEmpty.Get() {
@@ -74,11 +134,28 @@ func (s *peerAPIServer) hasFilesWaiting() bool {
 	for {
 		des, err := f.ReadDir(10)
 		for _, de := range des {
-			if strings.HasSuffix(de.Name(), partialSuffix) {
+			name := de.Name()
+			if strings.HasSuffix(name, partialSuffix) {
+				continue
+			}
+			if strings.HasSuffix(name, deletedSuffix) { // for Windows + tests
+				// After we're done looping over files, then try
+				// to delete this file. Don't do it proactively,
+				// as the OS may return "foo.jpg.deleted" before "foo.jpg"
+				// and we don't want to delete the ".deleted" file before
+				// enumerating to the "foo.jpg" file.
+				defer tryDeleteAgain(filepath.Join(s.rootDir, strings.TrimSuffix(name, deletedSuffix)))
 				continue
 			}
 			if de.Type().IsRegular() {
-				return true
+				_, err := os.Stat(filepath.Join(s.rootDir, name+deletedSuffix))
+				if os.IsNotExist(err) {
+					return true
+				}
+				if err == nil {
+					tryDeleteAgain(filepath.Join(s.rootDir, name))
+					continue
+				}
 			}
 		}
 		if err == io.EOF {
@@ -91,22 +168,25 @@ func (s *peerAPIServer) hasFilesWaiting() bool {
 	return false
 }
 
-// WaitingFile is a JSON-marshaled struct sent by the localapi to pick
-// up queued files.
-type WaitingFile struct {
-	Name string
-	Size int64
-}
-
-func (s *peerAPIServer) WaitingFiles() (ret []WaitingFile, err error) {
+// WaitingFiles returns the list of files that have been sent by a
+// peer that are waiting in the buffered "pick up" directory owned by
+// the Tailscale daemon.
+//
+// As a side effect, it also does any lazy deletion of files as
+// required by Windows.
+func (s *peerAPIServer) WaitingFiles() (ret []apitype.WaitingFile, err error) {
 	if s.rootDir == "" {
 		return nil, errors.New("peerapi disabled; no storage configured")
 	}
+	if s.directFileMode {
+		return nil, nil
+	}
 	f, err := os.Open(s.rootDir)
 	if err != nil {
 		return nil, err
 	}
 	defer f.Close()
+	var deleted map[string]bool // "foo.jpg" => true (if "foo.jpg.deleted" exists)
 	for {
 		des, err := f.ReadDir(10)
 		for _, de := range des {
@@ -114,12 +194,19 @@ func (s *peerAPIServer) WaitingFiles() (ret []WaitingFile, err error) {
 			if strings.HasSuffix(name, partialSuffix) {
 				continue
 			}
+			if strings.HasSuffix(name, deletedSuffix) { // for Windows + tests
+				if deleted == nil {
+					deleted = map[string]bool{}
+				}
+				deleted[strings.TrimSuffix(name, deletedSuffix)] = true
+				continue
+			}
 			if de.Type().IsRegular() {
 				fi, err := de.Info()
 				if err != nil {
 					continue
 				}
-				ret = append(ret, WaitingFile{
+				ret = append(ret, apitype.WaitingFile{
 					Name: filepath.Base(name),
 					Size: fi.Size(),
 				})
@@ -132,40 +219,130 @@ func (s *peerAPIServer) WaitingFiles() (ret []WaitingFile, err error) {
 			return nil, err
 		}
 	}
+	if len(deleted) > 0 {
+		// Filter out any return values "foo.jpg" where a
+		// "foo.jpg.deleted" marker file exists on disk.
+		all := ret
+		ret = ret[:0]
+		for _, wf := range all {
+			if !deleted[wf.Name] {
+				ret = append(ret, wf)
+			}
+		}
+		// And do some opportunistic deleting while we're here.
+		// Maybe Windows is done virus scanning the file we tried
+		// to delete a long time ago and will let us delete it now.
+		for name := range deleted {
+			tryDeleteAgain(filepath.Join(s.rootDir, name))
+		}
+	}
+	sort.Slice(ret, func(i, j int) bool { return ret[i].Name < ret[j].Name })
 	return ret, nil
 }
 
+// tryDeleteAgain tries to delete path (and path+deletedSuffix) after
+// it failed earlier.  This happens on Windows when various anti-virus
+// tools hook into filesystem operations and have the file open still
+// while we're trying to delete it. In that case we instead mark it as
+// deleted (writing a "foo.jpg.deleted" marker file), but then we
+// later try to clean them up.
+//
+// fullPath is the full path to the file without the deleted suffix.
+func tryDeleteAgain(fullPath string) {
+	if err := os.Remove(fullPath); err == nil || os.IsNotExist(err) {
+		os.Remove(fullPath + deletedSuffix)
+	}
+}
+
 func (s *peerAPIServer) DeleteFile(baseName string) error {
 	if s.rootDir == "" {
 		return errors.New("peerapi disabled; no storage configured")
 	}
+	if s.directFileMode {
+		return errors.New("deletes not allowed in direct mode")
+	}
 	path, ok := s.diskPath(baseName)
 	if !ok {
 		return errors.New("bad filename")
 	}
-	err := os.Remove(path)
-	if err != nil && !os.IsNotExist(err) {
-		return err
+	var bo *backoff.Backoff
+	logf := s.b.logf
+	t0 := time.Now()
+	for {
+		err := os.Remove(path)
+		if err != nil && !os.IsNotExist(err) {
+			err = redactErr(err)
+			// Put a retry loop around deletes on Windows. Windows
+			// file descriptor closes are effectively asynchronous,
+			// as a bunch of hooks run on/after close, and we can't
+			// necessarily delete the file for a while after close,
+			// as we need to wait for everybody to be done with
+			// it. (on Windows, unlike Unix, a file can't be deleted
+			// if it's open anywhere)
+			// So try a few times but ultimately just leave a
+			// "foo.jpg.deleted" marker file to note that it's
+			// deleted and we clean it up later.
+			if runtime.GOOS == "windows" {
+				if bo == nil {
+					bo = backoff.NewBackoff("delete-retry", logf, 1*time.Second)
+				}
+				if time.Since(t0) < 5*time.Second {
+					bo.BackOff(context.Background(), err)
+					continue
+				}
+				if err := touchFile(path + deletedSuffix); err != nil {
+					logf("peerapi: failed to leave deleted marker: %v", err)
+				}
+			}
+			logf("peerapi: failed to DeleteFile: %v", err)
+			return err
+		}
+		return nil
 	}
-	return nil
+}
+
+// redacted is a fake path name we use in errors, to avoid
+// accidentally logging actual filenames anywhere.
+const redacted = "redacted"
+
+func redactErr(err error) error {
+	if pe, ok := err.(*os.PathError); ok {
+		pe.Path = redacted
+	}
+	return err
+}
+
+func touchFile(path string) error {
+	f, err := os.OpenFile(path, os.O_RDWR|os.O_CREATE, 0666)
+	if err != nil {
+		return redactErr(err)
+	}
+	return f.Close()
 }
 
 func (s *peerAPIServer) OpenFile(baseName string) (rc io.ReadCloser, size int64, err error) {
 	if s.rootDir == "" {
 		return nil, 0, errors.New("peerapi disabled; no storage configured")
 	}
+	if s.directFileMode {
+		return nil, 0, errors.New("opens not allowed in direct mode")
+	}
 	path, ok := s.diskPath(baseName)
 	if !ok {
 		return nil, 0, errors.New("bad filename")
 	}
+	if fi, err := os.Stat(path + deletedSuffix); err == nil && fi.Mode().IsRegular() {
+		tryDeleteAgain(path)
+		return nil, 0, &fs.PathError{Op: "open", Path: redacted, Err: fs.ErrNotExist}
+	}
 	f, err := os.Open(path)
 	if err != nil {
-		return nil, 0, err
+		return nil, 0, redactErr(err)
 	}
 	fi, err := f.Stat()
 	if err != nil {
 		f.Close()
-		return nil, 0, err
+		return nil, 0, redactErr(err)
 	}
 	return f, fi.Size(), nil
 }
@@ -279,7 +456,6 @@ func (pln *peerAPIListener) serve() {
 			remoteAddr: ipp,
 			peerNode:   peerNode,
 			peerUser:   peerUser,
-			lb:         pln.lb,
 		}
 		httpServer := &http.Server{
 			Handler: h,
@@ -313,7 +489,6 @@ type peerAPIHandler struct {
 	isSelf     bool                // whether peerNode is owned by same user as this node
 	peerNode   *tailcfg.Node       // peerNode is who's making the request
 	peerUser   tailcfg.UserProfile // profile of peerNode
-	lb         *LocalBackend
 }
 
 func (h *peerAPIHandler) logf(format string, a ...interface{}) {
@@ -322,7 +497,11 @@ func (h *peerAPIHandler) logf(format string, a ...interface{}) {
 
 func (h *peerAPIHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	if strings.HasPrefix(r.URL.Path, "/v0/put/") {
-		h.put(w, r)
+		h.handlePeerPut(w, r)
+		return
+	}
+	if r.URL.Path == "/v0/goroutines" {
+		h.handleServeGoroutines(w, r)
 		return
 	}
 	who := h.peerUser.DisplayName
@@ -331,65 +510,179 @@ func (h *peerAPIHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 <body>
 <h1>Hello, %s (%v)</h1>
 This is my Tailscale device. Your device is %v.
-`, html.EscapeString(who), h.remoteAddr.IP, html.EscapeString(h.peerNode.ComputedName))
+`, html.EscapeString(who), h.remoteAddr.IP(), html.EscapeString(h.peerNode.ComputedName))
 
 	if h.isSelf {
 		fmt.Fprintf(w, "<p>You are the owner of this node.\n")
 	}
 }
 
-func (h *peerAPIHandler) put(w http.ResponseWriter, r *http.Request) {
+type incomingFile struct {
+	name        string // "foo.jpg"
+	started     time.Time
+	size        int64     // or -1 if unknown; never 0
+	w           io.Writer // underlying writer
+	ph          *peerAPIHandler
+	partialPath string // non-empty in direct mode
+
+	mu         sync.Mutex
+	copied     int64
+	done       bool
+	lastNotify time.Time
+}
+
+func (f *incomingFile) markAndNotifyDone() {
+	f.mu.Lock()
+	f.done = true
+	f.mu.Unlock()
+	b := f.ph.ps.b
+	b.sendFileNotify()
+}
+
+func (f *incomingFile) Write(p []byte) (n int, err error) {
+	n, err = f.w.Write(p)
+
+	b := f.ph.ps.b
+	var needNotify bool
+	defer func() {
+		if needNotify {
+			b.sendFileNotify()
+		}
+	}()
+	if n > 0 {
+		f.mu.Lock()
+		defer f.mu.Unlock()
+		f.copied += int64(n)
+		now := time.Now()
+		if f.lastNotify.IsZero() || now.Sub(f.lastNotify) > time.Second {
+			f.lastNotify = now
+			needNotify = true
+		}
+	}
+	return n, err
+}
+
+func (f *incomingFile) PartialFile() ipn.PartialFile {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return ipn.PartialFile{
+		Name:         f.name,
+		Started:      f.started,
+		DeclaredSize: f.size,
+		Received:     f.copied,
+		PartialPath:  f.partialPath,
+		Done:         f.done,
+	}
+}
+
+func (h *peerAPIHandler) handlePeerPut(w http.ResponseWriter, r *http.Request) {
 	if !h.isSelf {
 		http.Error(w, "not owner", http.StatusForbidden)
 		return
 	}
+	if !h.ps.b.hasCapFileSharing() {
+		http.Error(w, "file sharing not enabled by Tailscale admin", http.StatusForbidden)
+		return
+	}
 	if r.Method != "PUT" {
-		http.Error(w, "not method PUT", http.StatusMethodNotAllowed)
+		http.Error(w, "expected method PUT", http.StatusMethodNotAllowed)
 		return
 	}
 	if h.ps.rootDir == "" {
 		http.Error(w, "no rootdir", http.StatusInternalServerError)
 		return
 	}
-	baseName := path.Base(r.URL.Path)
+	rawPath := r.URL.EscapedPath()
+	suffix := strings.TrimPrefix(rawPath, "/v0/put/")
+	if suffix == rawPath {
+		http.Error(w, "misconfigured internals", 500)
+		return
+	}
+	if suffix == "" {
+		http.Error(w, "empty filename", 400)
+		return
+	}
+	if strings.Contains(suffix, "/") {
+		http.Error(w, "directories not supported", 400)
+		return
+	}
+	baseName, err := url.PathUnescape(suffix)
+	if err != nil {
+		http.Error(w, "bad path encoding", 400)
+		return
+	}
 	dstFile, ok := h.ps.diskPath(baseName)
 	if !ok {
 		http.Error(w, "bad filename", 400)
 		return
 	}
-	f, err := os.Create(dstFile)
+	t0 := time.Now()
+	// TODO(bradfitz): prevent same filename being sent by two peers at once
+	partialFile := dstFile + partialSuffix
+	f, err := os.Create(partialFile)
 	if err != nil {
-		h.logf("put Create error: %v", err)
+		h.logf("put Create error: %v", redactErr(err))
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 	var success bool
 	defer func() {
 		if !success {
-			os.Remove(dstFile)
+			os.Remove(partialFile)
 		}
 	}()
-	n, err := io.Copy(f, r.Body)
-	if err != nil {
-		f.Close()
-		h.logf("put Copy error: %v", err)
-		http.Error(w, err.Error(), http.StatusInternalServerError)
-		return
+	var finalSize int64
+	var inFile *incomingFile
+	if r.ContentLength != 0 {
+		inFile = &incomingFile{
+			name:    baseName,
+			started: time.Now(),
+			size:    r.ContentLength,
+			w:       f,
+			ph:      h,
+		}
+		if h.ps.directFileMode {
+			inFile.partialPath = partialFile
+		}
+		h.ps.b.registerIncomingFile(inFile, true)
+		defer h.ps.b.registerIncomingFile(inFile, false)
+		n, err := io.Copy(inFile, r.Body)
+		if err != nil {
+			err = redactErr(err)
+			f.Close()
+			h.logf("put Copy error: %v", err)
+			http.Error(w, err.Error(), http.StatusInternalServerError)
+			return
+		}
+		finalSize = n
 	}
-	if err := f.Close(); err != nil {
+	if err := redactErr(f.Close()); err != nil {
 		h.logf("put Close error: %v", err)
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
+	if h.ps.directFileMode {
+		if inFile != nil { // non-zero length; TODO: notify even for zero length
+			inFile.markAndNotifyDone()
+		}
+	} else {
+		if err := os.Rename(partialFile, dstFile); err != nil {
+			err = redactErr(err)
+			h.logf("put final rename: %v", err)
+			http.Error(w, err.Error(), http.StatusInternalServerError)
+			return
+		}
+	}
 
-	h.logf("put of %s from %v/%v", baseName, approxSize(n), h.remoteAddr.IP, h.peerNode.ComputedName)
+	d := time.Since(t0).Round(time.Second / 10)
+	h.logf("got put of %s in %v from %v/%v", approxSize(finalSize), d, h.remoteAddr.IP, h.peerNode.ComputedName)
 
 	// TODO: set modtime
 	// TODO: some real response
 	success = true
 	io.WriteString(w, "{}\n")
 	h.ps.knownEmpty.Set(false)
-	h.ps.b.send(ipn.Notify{}) // it will set FilesWaiting
+	h.ps.b.sendFileNotify()
 }
 
 func approxSize(n int64) string {
@@ -399,5 +692,21 @@ func approxSize(n int64) string {
 	if n <= 1<<20 {
 		return "<=1MB"
 	}
-	return fmt.Sprintf("~%dMB", n/1<<20)
+	return fmt.Sprintf("~%dMB", n>>20)
+}
+
+func (h *peerAPIHandler) handleServeGoroutines(w http.ResponseWriter, r *http.Request) {
+	if !h.isSelf {
+		http.Error(w, "not owner", http.StatusForbidden)
+		return
+	}
+	var buf []byte
+	for size := 4 << 10; size <= 2<<20; size *= 2 {
+		buf = make([]byte, size)
+		buf = buf[:runtime.Stack(buf, true)]
+		if len(buf) < size {
+			break
+		}
+	}
+	w.Write(buf)
 }

ipn/ipnlocal/peerapi_macios_ext.go

@@ -2,24 +2,25 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build darwin,redo ios,redo
+//go:build (darwin && ts_macext) || (ios && ts_macext)
+// +build darwin,ts_macext ios,ts_macext
 
 package ipnlocal
 
 import (
+	"errors"
 	"fmt"
-	"log"
 	"net"
-	"strings"
 	"syscall"
 
-	"golang.org/x/sys/unix"
 	"inet.af/netaddr"
 	"tailscale.com/net/interfaces"
+	"tailscale.com/net/netns"
 )
 
 func init() {
 	initListenConfig = initListenConfigNetworkExtension
+	peerDialControlFunc = peerDialControlFuncNetworkExtension
 }
 
 // initListenConfigNetworkExtension configures nc for listening on IP
@@ -30,25 +31,26 @@ func initListenConfigNetworkExtension(nc *net.ListenConfig, ip netaddr.IP, st *i
 	if !ok {
 		return fmt.Errorf("no interface with name %q", tunIfName)
 	}
-	nc.Control = func(network, address string, c syscall.RawConn) error {
-		var sockErr error
-		err := c.Control(func(fd uintptr) {
-
-			v6 := strings.Contains(address, "]:") || strings.HasSuffix(network, "6") // hacky test for v6
-			proto := unix.IPPROTO_IP
-			opt := unix.IP_BOUND_IF
-			if v6 {
-				proto = unix.IPPROTO_IPV6
-				opt = unix.IPV6_BOUND_IF
-			}
-
-			sockErr = unix.SetsockoptInt(int(fd), proto, opt, tunIf.Index)
-			log.Printf("peerapi: bind(%q, %q) on index %v = %v", network, address, tunIf.Index, sockErr)
-		})
-		if err != nil {
-			return err
-		}
-		return sockErr
-	}
-	return nil
+	return netns.SetListenConfigInterfaceIndex(nc, tunIf.Index)
+}
+
+func peerDialControlFuncNetworkExtension(b *LocalBackend) func(network, address string, c syscall.RawConn) error {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	st := b.prevIfState
+	pas := b.peerAPIServer
+	index := -1
+	if st != nil && pas != nil && pas.tunName != "" {
+		if tunIf, ok := st.Interface[pas.tunName]; ok {
+			index = tunIf.Index
+		}
+	}
+	var lc net.ListenConfig
+	netns.SetListenConfigInterfaceIndex(&lc, index)
+	return func(network, address string, c syscall.RawConn) error {
+		if index == -1 {
+			return errors.New("failed to find TUN interface to bind to")
+		}
+		return lc.Control(network, address, c)
+	}
 }

ipn/ipnlocal/peerapi_test.go

@@ -0,0 +1,573 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ipnlocal
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"io/fs"
+	"io/ioutil"
+	"math/rand"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"testing"
+
+	"tailscale.com/tailcfg"
+)
+
+type peerAPITestEnv struct {
+	ph     *peerAPIHandler
+	rr     *httptest.ResponseRecorder
+	logBuf bytes.Buffer
+}
+
+func (e *peerAPITestEnv) logf(format string, a ...interface{}) {
+	fmt.Fprintf(&e.logBuf, format, a...)
+}
+
+type check func(*testing.T, *peerAPITestEnv)
+
+func checks(vv ...check) []check { return vv }
+
+func httpStatus(wantStatus int) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		if res := e.rr.Result(); res.StatusCode != wantStatus {
+			t.Errorf("HTTP response code = %v; want %v", res.Status, wantStatus)
+		}
+	}
+}
+
+func bodyContains(sub string) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		if body := e.rr.Body.String(); !strings.Contains(body, sub) {
+			t.Errorf("HTTP response body does not contain %q; got: %s", sub, body)
+		}
+	}
+}
+
+func bodyNotContains(sub string) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		if body := e.rr.Body.String(); strings.Contains(body, sub) {
+			t.Errorf("HTTP response body unexpectedly contains %q; got: %s", sub, body)
+		}
+	}
+}
+
+func fileHasSize(name string, size int) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		root := e.ph.ps.rootDir
+		if root == "" {
+			t.Errorf("no rootdir; can't check whether %q has size %v", name, size)
+			return
+		}
+		path := filepath.Join(root, name)
+		if fi, err := os.Stat(path); err != nil {
+			t.Errorf("fileHasSize(%q, %v): %v", name, size, err)
+		} else if fi.Size() != int64(size) {
+			t.Errorf("file %q has size %v; want %v", name, fi.Size(), size)
+		}
+	}
+}
+
+func fileHasContents(name string, want string) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		root := e.ph.ps.rootDir
+		if root == "" {
+			t.Errorf("no rootdir; can't check contents of %q", name)
+			return
+		}
+		path := filepath.Join(root, name)
+		got, err := ioutil.ReadFile(path)
+		if err != nil {
+			t.Errorf("fileHasContents: %v", err)
+			return
+		}
+		if string(got) != want {
+			t.Errorf("file contents = %q; want %q", got, want)
+		}
+	}
+}
+
+func hexAll(v string) string {
+	var sb strings.Builder
+	for i := 0; i < len(v); i++ {
+		fmt.Fprintf(&sb, "%%%02x", v[i])
+	}
+	return sb.String()
+}
+
+func TestHandlePeerAPI(t *testing.T) {
+	tests := []struct {
+		name       string
+		isSelf     bool // the peer sending the request is owned by us
+		capSharing bool // self node has file sharing capabilty
+		omitRoot   bool // don't configure
+		req        *http.Request
+		checks     []check
+	}{
+		{
+			name:       "not_peer_api",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("GET", "/", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("This is my Tailscale device."),
+				bodyContains("You are the owner of this node."),
+			),
+		},
+		{
+			name:       "not_peer_api_not_owner",
+			isSelf:     false,
+			capSharing: true,
+			req:        httptest.NewRequest("GET", "/", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("This is my Tailscale device."),
+				bodyNotContains("You are the owner of this node."),
+			),
+		},
+		{
+			name:   "peer_api_goroutines_deny",
+			isSelf: false,
+			req:    httptest.NewRequest("GET", "/v0/goroutines", nil),
+			checks: checks(httpStatus(403)),
+		},
+		{
+			name:   "peer_api_goroutines",
+			isSelf: true,
+			req:    httptest.NewRequest("GET", "/v0/goroutines", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("ServeHTTP"),
+			),
+		},
+		{
+			name:       "reject_non_owner_put",
+			isSelf:     false,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(http.StatusForbidden),
+				bodyContains("not owner"),
+			),
+		},
+		{
+			name:       "owner_without_cap",
+			isSelf:     true,
+			capSharing: false,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(http.StatusForbidden),
+				bodyContains("file sharing not enabled by Tailscale admin"),
+			),
+		},
+		{
+			name:       "owner_with_cap_no_rootdir",
+			omitRoot:   true,
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(http.StatusInternalServerError),
+				bodyContains("no rootdir"),
+			),
+		},
+		{
+			name:       "bad_method",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("POST", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(405),
+				bodyContains("expected method PUT"),
+			),
+		},
+		{
+			name:       "put_zero_length",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasSize("foo", 0),
+				fileHasContents("foo", ""),
+			),
+		},
+		{
+			name:       "put_non_zero_length_content_length",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", strings.NewReader("contents")),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasSize("foo", len("contents")),
+				fileHasContents("foo", "contents"),
+			),
+		},
+		{
+			name:       "put_non_zero_length_chunked",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", struct{ io.Reader }{strings.NewReader("contents")}),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasSize("foo", len("contents")),
+				fileHasContents("foo", "contents"),
+			),
+		},
+		{
+			name:       "bad_filename_partial",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo.partial", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_deleted",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo.deleted", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_dot",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/.", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_empty",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("empty filename"),
+			),
+		},
+		{
+			name:       "bad_filename_slash",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo/bar", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("directories not supported"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_dot",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("."), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_slash",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("/"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_backslash",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("\\"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_dotdot",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll(".."), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_dotdot_out",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("foo/../../../../../etc/passwd"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_spaces_and_caps",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("Foo Bar.dat"), strings.NewReader("baz")),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasContents("Foo Bar.dat", "baz"),
+			),
+		},
+		{
+			name:       "put_unicode",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("Томас и его друзья.mp3"), strings.NewReader("главный озорник")),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasContents("Томас и его друзья.mp3", "главный озорник"),
+			),
+		},
+		{
+			name:       "put_invalid_utf8",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+(hexAll("😜")[:3]), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_null",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/%00", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_non_printable",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/%01", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_colon",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("nul:"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_surrounding_whitespace",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll(" foo "), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var e peerAPITestEnv
+			lb := &LocalBackend{
+				logf:           e.logf,
+				capFileSharing: tt.capSharing,
+			}
+			e.ph = &peerAPIHandler{
+				isSelf: tt.isSelf,
+				peerNode: &tailcfg.Node{
+					ComputedName: "some-peer-name",
+				},
+				ps: &peerAPIServer{
+					b: lb,
+				},
+			}
+			var rootDir string
+			if !tt.omitRoot {
+				rootDir = t.TempDir()
+				e.ph.ps.rootDir = rootDir
+			}
+			e.rr = httptest.NewRecorder()
+			e.ph.ServeHTTP(e.rr, tt.req)
+			for _, f := range tt.checks {
+				f(t, &e)
+			}
+			if t.Failed() && rootDir != "" {
+				t.Logf("Contents of %s:", rootDir)
+				des, _ := fs.ReadDir(os.DirFS(rootDir), ".")
+				for _, de := range des {
+					fi, err := de.Info()
+					if err != nil {
+						t.Log(err)
+					} else {
+						t.Logf("  %v %5d %s", fi.Mode(), fi.Size(), de.Name())
+					}
+				}
+			}
+		})
+	}
+}
+
+// Windows likes to hold on to file descriptors for some indeterminate
+// amount of time after you close them and not let you delete them for
+// a bit. So test that we work around that sufficiently.
+func TestFileDeleteRace(t *testing.T) {
+	dir := t.TempDir()
+	ps := &peerAPIServer{
+		b: &LocalBackend{
+			logf:           t.Logf,
+			capFileSharing: true,
+		},
+		rootDir: dir,
+	}
+	ph := &peerAPIHandler{
+		isSelf: true,
+		peerNode: &tailcfg.Node{
+			ComputedName: "some-peer-name",
+		},
+		ps: ps,
+	}
+	buf := make([]byte, 2<<20)
+	for i := 0; i < 30; i++ {
+		rr := httptest.NewRecorder()
+		ph.ServeHTTP(rr, httptest.NewRequest("PUT", "/v0/put/foo.txt", bytes.NewReader(buf[:rand.Intn(len(buf))])))
+		if res := rr.Result(); res.StatusCode != 200 {
+			t.Fatal(res.Status)
+		}
+		wfs, err := ps.WaitingFiles()
+		if err != nil {
+			t.Fatal(err)
+		}
+		if len(wfs) != 1 {
+			t.Fatalf("waiting files = %d; want 1", len(wfs))
+		}
+
+		if err := ps.DeleteFile("foo.txt"); err != nil {
+			t.Fatal(err)
+		}
+		wfs, err = ps.WaitingFiles()
+		if err != nil {
+			t.Fatal(err)
+		}
+		if len(wfs) != 0 {
+			t.Fatalf("waiting files = %d; want 0", len(wfs))
+		}
+	}
+}
+
+// Tests "foo.jpg.deleted" marks (for Windows).
+func TestDeletedMarkers(t *testing.T) {
+	dir := t.TempDir()
+	ps := &peerAPIServer{
+		b: &LocalBackend{
+			logf:           t.Logf,
+			capFileSharing: true,
+		},
+		rootDir: dir,
+	}
+
+	nothingWaiting := func() {
+		t.Helper()
+		ps.knownEmpty.Set(false)
+		if ps.hasFilesWaiting() {
+			t.Fatal("unexpected files waiting")
+		}
+	}
+	touch := func(base string) {
+		t.Helper()
+		if err := touchFile(filepath.Join(dir, base)); err != nil {
+			t.Fatal(err)
+		}
+	}
+	wantEmptyTempDir := func() {
+		t.Helper()
+		if fis, err := ioutil.ReadDir(dir); err != nil {
+			t.Fatal(err)
+		} else if len(fis) > 0 && runtime.GOOS != "windows" {
+			for _, fi := range fis {
+				t.Errorf("unexpected file in tempdir: %q", fi.Name())
+			}
+		}
+	}
+
+	nothingWaiting()
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	nothingWaiting()
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	touch("foo.jpg")
+	nothingWaiting()
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	touch("foo.jpg")
+	wf, err := ps.WaitingFiles()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(wf) != 0 {
+		t.Fatalf("WaitingFiles = %d; want 0", len(wf))
+	}
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	touch("foo.jpg")
+	if rc, _, err := ps.OpenFile("foo.jpg"); err == nil {
+		rc.Close()
+		t.Fatal("unexpected foo.jpg open")
+	}
+	wantEmptyTempDir()
+
+	// And verify basics still work in non-deleted cases.
+	touch("foo.jpg")
+	touch("bar.jpg.deleted")
+	if wf, err := ps.WaitingFiles(); err != nil {
+		t.Error(err)
+	} else if len(wf) != 1 {
+		t.Errorf("WaitingFiles = %d; want 1", len(wf))
+	} else if wf[0].Name != "foo.jpg" {
+		t.Errorf("unexpected waiting file %+v", wf[0])
+	}
+	if rc, _, err := ps.OpenFile("foo.jpg"); err != nil {
+		t.Fatal(err)
+	} else {
+		rc.Close()
+	}
+
+}

ipn/ipnlocal/state_test.go

@@ -0,0 +1,919 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ipnlocal
+
+import (
+	"context"
+	"sync"
+	"testing"
+	"time"
+
+	qt "github.com/frankban/quicktest"
+
+	"tailscale.com/control/controlclient"
+	"tailscale.com/ipn"
+	"tailscale.com/syncs"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/empty"
+	"tailscale.com/types/logger"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/persist"
+	"tailscale.com/types/wgkey"
+	"tailscale.com/wgengine"
+)
+
+// notifyThrottler receives notifications from an ipn.Backend, blocking
+// (with eventual timeout and t.Fatal) if there are too many and complaining
+// (also with t.Fatal) if they are too few.
+type notifyThrottler struct {
+	t *testing.T
+
+	// ch gets replaced frequently. Lock the mutex before getting or
+	// setting it, but not while waiting on it.
+	mu sync.Mutex
+	ch chan ipn.Notify
+}
+
+// expect tells the throttler to expect count upcoming notifications.
+func (nt *notifyThrottler) expect(count int) {
+	nt.mu.Lock()
+	nt.ch = make(chan ipn.Notify, count)
+	nt.mu.Unlock()
+}
+
+// put adds one notification into the throttler's queue.
+func (nt *notifyThrottler) put(n ipn.Notify) {
+	nt.mu.Lock()
+	ch := nt.ch
+	nt.mu.Unlock()
+
+	select {
+	case ch <- n:
+		return
+	default:
+		nt.t.Fatalf("put: channel full: %v", n)
+	}
+}
+
+// drain pulls the notifications out of the queue, asserting that there are
+// exactly count notifications that have been put so far.
+func (nt *notifyThrottler) drain(count int) []ipn.Notify {
+	nt.mu.Lock()
+	ch := nt.ch
+	nt.mu.Unlock()
+
+	nn := []ipn.Notify{}
+	for i := 0; i < count; i++ {
+		select {
+		case n := <-ch:
+			nn = append(nn, n)
+		case <-time.After(6 * time.Second):
+			nt.t.Fatalf("drain: channel empty after %d/%d", i, count)
+		}
+	}
+
+	// no more notifications expected
+	close(ch)
+
+	return nn
+}
+
+// mockControl is a mock implementation of controlclient.Client.
+// Much of the backend state machine depends on callbacks and state
+// in the controlclient.Client, so by controlling it, we can check that
+// the state machine works as expected.
+type mockControl struct {
+	opts       controlclient.Options
+	logf       logger.Logf
+	statusFunc func(controlclient.Status)
+
+	mu          sync.Mutex
+	calls       []string
+	authBlocked bool
+	persist     persist.Persist
+	machineKey  wgkey.Private
+}
+
+func newMockControl() *mockControl {
+	return &mockControl{
+		calls:       []string{},
+		authBlocked: true,
+	}
+}
+
+func (cc *mockControl) SetStatusFunc(fn func(controlclient.Status)) {
+	cc.statusFunc = fn
+}
+
+func (cc *mockControl) populateKeys() (newKeys bool) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	if cc.machineKey.IsZero() {
+		cc.logf("Copying machineKey.")
+		cc.machineKey, _ = cc.opts.GetMachinePrivateKey()
+		newKeys = true
+	}
+
+	if cc.persist.PrivateNodeKey.IsZero() {
+		cc.logf("Generating a new nodekey.")
+		cc.persist.OldPrivateNodeKey = cc.persist.PrivateNodeKey
+		cc.persist.PrivateNodeKey, _ = wgkey.NewPrivate()
+		newKeys = true
+	}
+
+	return newKeys
+}
+
+// send publishes a controlclient.Status notification upstream.
+// (In our tests here, upstream is the ipnlocal.Local instance.)
+func (cc *mockControl) send(err error, url string, loginFinished bool, nm *netmap.NetworkMap) {
+	if cc.statusFunc != nil {
+		s := controlclient.Status{
+			URL:     url,
+			NetMap:  nm,
+			Persist: &cc.persist,
+		}
+		if err != nil {
+			s.Err = err.Error()
+		}
+		if loginFinished {
+			s.LoginFinished = &empty.Message{}
+		} else if url == "" && err == nil && nm == nil {
+			s.LogoutFinished = &empty.Message{}
+		}
+		cc.statusFunc(s)
+	}
+}
+
+// called records that a particular function name was called.
+func (cc *mockControl) called(s string) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	cc.calls = append(cc.calls, s)
+}
+
+// getCalls returns the list of functions that have been called since the
+// last time getCalls was run.
+func (cc *mockControl) getCalls() []string {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	r := cc.calls
+	cc.calls = []string{}
+	return r
+}
+
+// setAuthBlocked changes the return value of AuthCantContinue.
+// Auth is blocked if you haven't called Login, the control server hasn't
+// provided an auth URL, or it has provided an auth URL and you haven't
+// visited it yet.
+func (cc *mockControl) setAuthBlocked(blocked bool) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	cc.authBlocked = blocked
+}
+
+// Shutdown disconnects the client.
+//
+// Note that in a normal controlclient, Shutdown would be the last thing you
+// do before discarding the object. In this mock, we don't actually discard
+// the object, but if you see a call to Shutdown, you should always see a
+// call to New right after it, if the object continues to be used.
+// (Note that "New" is the ccGen function here; it means ipn.Backend wanted
+// to create an entirely new controlclient.)
+func (cc *mockControl) Shutdown() {
+	cc.logf("Shutdown")
+	cc.called("Shutdown")
+}
+
+// Login starts a login process.
+// Note that in this mock, we don't automatically generate notifications
+// about the progress of the login operation. You have to call setAuthBlocked()
+// and send() as required by the test.
+func (cc *mockControl) Login(t *tailcfg.Oauth2Token, flags controlclient.LoginFlags) {
+	cc.logf("Login token=%v flags=%v", t, flags)
+	cc.called("Login")
+	newKeys := cc.populateKeys()
+
+	interact := (flags & controlclient.LoginInteractive) != 0
+	cc.logf("Login: interact=%v newKeys=%v", interact, newKeys)
+	cc.setAuthBlocked(interact || newKeys)
+}
+
+func (cc *mockControl) StartLogout() {
+	cc.logf("StartLogout")
+	cc.called("StartLogout")
+}
+
+func (cc *mockControl) Logout(ctx context.Context) error {
+	cc.logf("Logout")
+	cc.called("Logout")
+	return nil
+}
+
+func (cc *mockControl) SetPaused(paused bool) {
+	cc.logf("SetPaused=%v", paused)
+	if paused {
+		cc.called("pause")
+	} else {
+		cc.called("unpause")
+	}
+}
+
+func (cc *mockControl) AuthCantContinue() bool {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	return cc.authBlocked
+}
+
+func (cc *mockControl) SetHostinfo(hi *tailcfg.Hostinfo) {
+	cc.logf("SetHostinfo: %v", *hi)
+	cc.called("SetHostinfo")
+}
+
+func (cc *mockControl) SetNetInfo(ni *tailcfg.NetInfo) {
+	cc.called("SetNetinfo")
+	cc.logf("SetNetInfo: %v", *ni)
+	cc.called("SetNetInfo")
+}
+
+func (cc *mockControl) UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) {
+	// validate endpoint information here?
+	cc.logf("UpdateEndpoints: lp=%v ep=%v", localPort, endpoints)
+	cc.called("UpdateEndpoints")
+}
+
+func (*mockControl) SetDNS(context.Context, *tailcfg.SetDNSRequest) error {
+	panic("unexpected SetDNS call")
+}
+
+// A very precise test of the sequence of function calls generated by
+// ipnlocal.Local into its controlclient instance, and the events it
+// produces upstream into the UI.
+//
+// [apenwarr] Normally I'm not a fan of "mock" style tests, but the precise
+// sequence of this state machine is so important for writing our multiple
+// frontends, that it's worth validating it all in one place.
+//
+// Any changes that affect this test will most likely require carefully
+// re-testing all our GUIs (and the CLI) to make sure we didn't break
+// anything.
+//
+// Note also that this test doesn't have any timers, goroutines, or duplicate
+// detection. It expects messages to be produced in exactly the right order,
+// with no duplicates, without doing network activity (other than through
+// controlclient, which we fake, so there's no network activity there either).
+//
+// TODO: A few messages that depend on magicsock (which actually might have
+// network delays) are just ignored for now, which makes the test
+// predictable, but maybe a bit less thorough. This is more of an overall
+// state machine test than a test of the wgengine+magicsock integration.
+func TestStateMachine(t *testing.T) {
+	c := qt.New(t)
+
+	logf := t.Logf
+	store := new(testStateStorage)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, 0)
+	if err != nil {
+		t.Fatalf("NewFakeUserspaceEngine: %v", err)
+	}
+
+	cc := newMockControl()
+	b, err := NewLocalBackend(logf, "logid", store, e)
+	if err != nil {
+		t.Fatalf("NewLocalBackend: %v", err)
+	}
+	b.SetControlClientGetterForTesting(func(opts controlclient.Options) (controlclient.Client, error) {
+		cc.mu.Lock()
+		cc.opts = opts
+		cc.logf = opts.Logf
+		cc.authBlocked = true
+		cc.persist = cc.opts.Persist
+		cc.mu.Unlock()
+
+		cc.logf("ccGen: new mockControl.")
+		cc.called("New")
+		return cc, nil
+	})
+
+	notifies := &notifyThrottler{t: t}
+	notifies.expect(0)
+
+	b.SetNotifyCallback(func(n ipn.Notify) {
+		if n.State != nil ||
+			n.Prefs != nil ||
+			n.BrowseToURL != nil ||
+			n.LoginFinished != nil {
+			logf("\n%v\n\n", n)
+			notifies.put(n)
+		} else {
+			logf("\n(ignored) %v\n\n", n)
+		}
+	})
+
+	// Check that it hasn't called us right away.
+	// The state machine should be idle until we call Start().
+	c.Assert(cc.getCalls(), qt.HasLen, 0)
+
+	// Start the state machine.
+	// Since !WantRunning by default, it'll create a controlclient,
+	// but not ask it to do anything yet.
+	t.Logf("\n\nStart")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// BUG: strictly, it should pause, not unpause, here, since !WantRunning.
+		c.Assert([]string{"New", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		prefs := *nn[0].Prefs
+		// Note: a totally fresh system has Prefs.LoggedOut=false by
+		// default. We are logged out, but not because the user asked
+		// for it, so it doesn't count as Prefs.LoggedOut==true.
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Restart the state machine.
+	// It's designed to handle frontends coming and going sporadically.
+	// Make the sure the restart not only works, but generates the same
+	// events as the first time, so UIs always know what to expect.
+	t.Logf("\n\nStart2")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// BUG: strictly, it should pause, not unpause, here, since !WantRunning.
+		c.Assert([]string{"Shutdown", "unpause", "New", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Start non-interactive login with no token.
+	// This will ask controlclient to start its own Login() process,
+	// then wait for us to respond.
+	t.Logf("\n\nLogin (noninteractive)")
+	notifies.expect(0)
+	b.Login(nil)
+	{
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"Login"})
+		notifies.drain(0)
+		// Note: WantRunning isn't true yet. It'll switch to true
+		// after a successful login finishes.
+		// (This behaviour is needed so that b.Login() won't
+		// start connecting to an old account right away, if one
+		// exists when you launch another login.)
+	}
+
+	// Attempted non-interactive login with no key; indicate that
+	// the user needs to visit a login URL.
+	t.Logf("\n\nLogin (url response)")
+	notifies.expect(1)
+	url1 := "http://localhost:1/1"
+	cc.send(nil, url1, false, nil)
+	{
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"unpause"})
+
+		// ...but backend eats that notification, because the user
+		// didn't explicitly request interactive login yet, and
+		// we're already in NeedsLogin state.
+		nn := notifies.drain(1)
+
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+	}
+
+	// Now we'll try an interactive login.
+	// Since we provided an interactive URL earlier, this shouldn't
+	// ask control to do anything. Instead backend will emit an event
+	// indicating that the UI should browse to the given URL.
+	t.Logf("\n\nLogin (interactive)")
+	notifies.expect(1)
+	b.StartLoginInteractive()
+	{
+		nn := notifies.drain(1)
+		// BUG: UpdateEndpoints shouldn't be called yet.
+		// We're still not logged in so there's nothing we can do
+		// with it. (And empirically, it's providing an empty list
+		// of endpoints.)
+		c.Assert([]string{"UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(url1, qt.Equals, *nn[0].BrowseToURL)
+	}
+
+	// Sometimes users press the Login button again, in the middle of
+	// a login sequence. For example, they might have closed their
+	// browser window without logging in, or they waited too long and
+	// the login URL expired. If they start another interactive login,
+	// we must always get a *new* login URL first.
+	t.Logf("\n\nLogin2 (interactive)")
+	notifies.expect(0)
+	b.StartLoginInteractive()
+	{
+		notifies.drain(0)
+		// backend asks control for another login sequence
+		c.Assert([]string{"Login"}, qt.DeepEquals, cc.getCalls())
+	}
+
+	// Provide a new interactive login URL.
+	t.Logf("\n\nLogin2 (url response)")
+	notifies.expect(1)
+	url2 := "http://localhost:1/2"
+	cc.send(nil, url2, false, nil)
+	{
+		// BUG: UpdateEndpoints again, this is getting silly.
+		c.Assert([]string{"UpdateEndpoints", "unpause", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		// This time, backend should emit it to the UI right away,
+		// because the UI is anxiously awaiting a new URL to visit.
+		nn := notifies.drain(1)
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(url2, qt.Equals, *nn[0].BrowseToURL)
+	}
+
+	// Pretend that the interactive login actually happened.
+	// Controlclient always sends the netmap and LoginFinished at the
+	// same time.
+	// The backend should propagate this upward for the UI.
+	t.Logf("\n\nLoginFinished")
+	notifies.expect(3)
+	cc.setAuthBlocked(false)
+	cc.persist.LoginName = "user1"
+	cc.send(nil, "", true, &netmap.NetworkMap{})
+	{
+		nn := notifies.drain(3)
+		// BUG: still too soon for UpdateEndpoints.
+		//
+		// Arguably it makes sense to unpause now, since the machine
+		// authorization status is part of the netmap.
+		//
+		// BUG: backend unblocks wgengine at this point, even though
+		// our machine key is not authorized. It probably should
+		// wait until it gets into Starting.
+		// TODO: (Currently this test doesn't detect that bug, but
+		// it's visible in the logs)
+		c.Assert([]string{"unpause", "unpause", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[2].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs.Persist.LoginName, qt.Equals, "user1")
+		c.Assert(ipn.NeedsMachineAuth, qt.Equals, *nn[2].State)
+	}
+
+	// Pretend that the administrator has authorized our machine.
+	t.Logf("\n\nMachineAuthorized")
+	notifies.expect(1)
+	// BUG: the real controlclient sends LoginFinished with every
+	// notification while it's in StateAuthenticated, but not StateSynced.
+	// It should send it exactly once, or every time we're authenticated,
+	// but the current code is brittle.
+	// (ie. I suspect it would be better to change false->true in send()
+	// below, and do the same in the real controlclient.)
+	cc.send(nil, "", false, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(1)
+		c.Assert([]string{"unpause", "unpause", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+	}
+
+	// TODO: add a fake DERP server to our fake netmap, so we can
+	// transition to the Running state here.
+
+	// TODO: test what happens when the admin forcibly deletes our key.
+	// (ie. unsolicited logout)
+
+	// TODO: test what happens when our key expires, client side.
+	// (and when it gets close to expiring)
+
+	// The user changes their preference to !WantRunning.
+	t.Logf("\n\nWantRunning -> false")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: false},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"pause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+	}
+
+	// The user changes their preference to WantRunning after all.
+	t.Logf("\n\nWantRunning -> true")
+	store.awaitWrite()
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: true},
+	})
+	{
+		nn := notifies.drain(2)
+		// BUG: UpdateEndpoints isn't needed here.
+		// BUG: Login isn't needed here. We never logged out.
+		c.Assert([]string{"Login", "unpause", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+		c.Assert(store.sawWrite(), qt.IsTrue)
+	}
+
+	// Test the fast-path frontend reconnection.
+	// This one is very finicky, so we have to force State==Running
+	// or it won't use the fast path.
+	// TODO: actually get to State==Running, rather than cheating.
+	//  That'll require spinning up a fake DERP server and putting it in
+	//  the netmap.
+	t.Logf("\n\nFastpath Start()")
+	notifies.expect(1)
+	b.state = ipn.Running
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		nn := notifies.drain(1)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[0].NetMap, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+	}
+
+	// undo the state hack above.
+	b.state = ipn.Starting
+
+	// User wants to logout.
+	store.awaitWrite()
+	t.Logf("\n\nLogout (async)")
+	notifies.expect(2)
+	b.Logout()
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"pause", "StartLogout", "pause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsTrue)
+		c.Assert(nn[1].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.Stopped, qt.Equals, b.State())
+		c.Assert(store.sawWrite(), qt.IsTrue)
+	}
+
+	// Let's make the logout succeed.
+	t.Logf("\n\nLogout (async) - succeed")
+	notifies.expect(1)
+	cc.setAuthBlocked(true)
+	cc.send(nil, "", false, nil)
+	{
+		nn := notifies.drain(1)
+		c.Assert([]string{"unpause", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[0].State)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// A second logout should do nothing, since the prefs haven't changed.
+	t.Logf("\n\nLogout2 (async)")
+	notifies.expect(0)
+	b.Logout()
+	{
+		notifies.drain(0)
+		// BUG: the backend has already called StartLogout, and we're
+		// still logged out. So it shouldn't call it again.
+		c.Assert([]string{"StartLogout", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Let's acknowledge the second logout too.
+	t.Logf("\n\nLogout2 (async) - succeed")
+	notifies.expect(0)
+	cc.setAuthBlocked(true)
+	cc.send(nil, "", false, nil)
+	{
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"unpause", "unpause"})
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Try the synchronous logout feature.
+	t.Logf("\n\nLogout3 (sync)")
+	notifies.expect(0)
+	b.LogoutSync(context.Background())
+	// NOTE: This returns as soon as cc.Logout() returns, which is okay
+	// I guess, since that's supposed to be synchronous.
+	{
+		notifies.drain(0)
+		c.Assert([]string{"Logout", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Generate the third logout event.
+	t.Logf("\n\nLogout3 (sync) - succeed")
+	notifies.expect(0)
+	cc.setAuthBlocked(true)
+	cc.send(nil, "", false, nil)
+	{
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"unpause", "unpause"})
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Shut down the backend.
+	t.Logf("\n\nShutdown")
+	notifies.expect(0)
+	b.Shutdown()
+	{
+		notifies.drain(0)
+		// BUG: I expect a transition to ipn.NoState here.
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"Shutdown"})
+	}
+
+	// Oh, you thought we were done? Ha! Now we have to test what
+	// happens if the user exits and restarts while logged out.
+	// Note that it's explicitly okay to call b.Start() over and over
+	// again, every time the frontend reconnects.
+
+	// TODO: test user switching between statekeys.
+
+	// The frontend restarts!
+	t.Logf("\n\nStart3")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// BUG: We already called Shutdown(), no need to do it again.
+		// BUG: Way too soon for UpdateEndpoints.
+		// BUG: don't unpause because we're not logged in.
+		c.Assert([]string{"Shutdown", "unpause", "New", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsTrue)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Let's break the rules a little. Our control server accepts
+	// your invalid login attempt, with no need for an interactive login.
+	// (This simulates an admin reviving a key that you previously
+	// disabled.)
+	t.Logf("\n\nLoginFinished3")
+	notifies.expect(3)
+	cc.setAuthBlocked(false)
+	cc.persist.LoginName = "user2"
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(3)
+		c.Assert([]string{"unpause", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[2].State, qt.Not(qt.IsNil))
+		// Prefs after finishing the login, so LoginName updated.
+		c.Assert(nn[1].Prefs.Persist.LoginName, qt.Equals, "user2")
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[1].Prefs.WantRunning, qt.IsTrue)
+		c.Assert(ipn.Starting, qt.Equals, *nn[2].State)
+	}
+
+	// Now we've logged in successfully. Let's disconnect.
+	t.Logf("\n\nWantRunning -> false")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: false},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"pause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsFalse)
+	}
+
+	// One more restart, this time with a valid key, but WantRunning=false.
+	t.Logf("\n\nStart4")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// NOTE: cc.Shutdown() is correct here, since we didn't call
+		// b.Shutdown() explicitly ourselves.
+		// BUG: UpdateEndpoints should be called here since we're not WantRunning.
+		// Note: unpause happens because ipn needs to get at least one netmap
+		//  on startup, otherwise UIs can't show the node list, login
+		//  name, etc when in state ipn.Stopped.
+		//  Arguably they shouldn't try. But they currently do.
+		nn := notifies.drain(2)
+		c.Assert([]string{"Shutdown", "unpause", "New", "UpdateEndpoints", "Login", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(ipn.Stopped, qt.Equals, *nn[1].State)
+	}
+
+	// When logged in but !WantRunning, ipn leaves us unpaused to retrieve
+	// the first netmap. Simulate that netmap being received, after which
+	// it should pause us, to avoid wasting CPU retrieving unnecessarily
+	// additional netmap updates.
+	//
+	// TODO: really the various GUIs and prefs should be refactored to
+	//  not require the netmap structure at all when starting while
+	//  !WantRunning. That would remove the need for this (or contacting
+	//  the control server at all when stopped).
+	t.Logf("\n\nStart4 -> netmap")
+	notifies.expect(0)
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		notifies.drain(0)
+		c.Assert([]string{"pause", "pause"}, qt.DeepEquals, cc.getCalls())
+	}
+
+	// Request connection.
+	// The state machine didn't call Login() earlier, so now it needs to.
+	t.Logf("\n\nWantRunning4 -> true")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: true},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"Login", "unpause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+	}
+
+	// Disconnect.
+	t.Logf("\n\nStop")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: false},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"pause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+	}
+
+	// We want to try logging in as a different user, while Stopped.
+	// First, start the login process (without logging out first).
+	t.Logf("\n\nLoginDifferent")
+	notifies.expect(1)
+	b.StartLoginInteractive()
+	url3 := "http://localhost:1/3"
+	cc.send(nil, url3, false, nil)
+	{
+		nn := notifies.drain(1)
+		// It might seem like WantRunning should switch to true here,
+		// but that would be risky since we already have a valid
+		// user account. It might try to reconnect to the old account
+		// before the new one is ready. So no change yet.
+		//
+		// Because the login hasn't yet completed, the old login
+		// is still valid, so it's correct that we stay paused.
+		c.Assert([]string{"Login", "pause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(*nn[0].BrowseToURL, qt.Equals, url3)
+	}
+
+	// Now, let's complete the interactive login, using a different
+	// user account than before. WantRunning changes to true after an
+	// interactive login, so we end up unpaused.
+	t.Logf("\n\nLoginDifferent URL visited")
+	notifies.expect(3)
+	cc.persist.LoginName = "user3"
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(3)
+		// BUG: pause() being called here is a bad sign.
+		//  It means that either the state machine ran at least once
+		//  with the old netmap, or it ran with the new login+netmap
+		//  and !WantRunning. But since it's a fresh and successful
+		//  new login, WantRunning is true, so there was never a
+		//  reason to pause().
+		c.Assert([]string{"pause", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[2].State, qt.Not(qt.IsNil))
+		// Prefs after finishing the login, so LoginName updated.
+		c.Assert(nn[1].Prefs.Persist.LoginName, qt.Equals, "user3")
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[1].Prefs.WantRunning, qt.IsTrue)
+		c.Assert(ipn.Starting, qt.Equals, *nn[2].State)
+	}
+
+	// The last test case is the most common one: restarting when both
+	// logged in and WantRunning.
+	t.Logf("\n\nStart5")
+	notifies.expect(1)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// NOTE: cc.Shutdown() is correct here, since we didn't call
+		// b.Shutdown() ourselves.
+		c.Assert([]string{"Shutdown", "unpause", "New", "UpdateEndpoints", "Login", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(1)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsTrue)
+		c.Assert(ipn.NoState, qt.Equals, b.State())
+	}
+
+	// Control server accepts our valid key from before.
+	t.Logf("\n\nLoginFinished5")
+	notifies.expect(1)
+	cc.setAuthBlocked(false)
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(1)
+		c.Assert([]string{"unpause", "unpause"}, qt.DeepEquals, cc.getCalls())
+		// NOTE: No LoginFinished message since no interactive
+		// login was needed.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+		// NOTE: No prefs change this time. WantRunning stays true.
+		// We were in Starting in the first place, so that doesn't
+		// change either.
+		c.Assert(ipn.Starting, qt.Equals, b.State())
+	}
+}
+
+type testStateStorage struct {
+	mem     ipn.MemoryStore
+	written syncs.AtomicBool
+}
+
+func (s *testStateStorage) ReadState(id ipn.StateKey) ([]byte, error) {
+	return s.mem.ReadState(id)
+}
+
+func (s *testStateStorage) WriteState(id ipn.StateKey, bs []byte) error {
+	s.written.Set(true)
+	return s.mem.WriteState(id, bs)
+}
+
+// awaitWrite clears the "I've seen writes" bit, in prep for a future
+// call to sawWrite to see if a write arrived.
+func (s *testStateStorage) awaitWrite() { s.written.Set(false) }
+
+// sawWrite reports whether there's been a WriteState call since the most
+// recent awaitWrite call.
+func (s *testStateStorage) sawWrite() bool {
+	v := s.written.Get()
+	s.awaitWrite()
+	return v
+}

ipn/ipnserver/server.go

@@ -6,7 +6,9 @@ package ipnserver
 
 import (
 	"bufio"
+	"bytes"
 	"context"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"io"
@@ -22,7 +24,6 @@ import (
 	"strconv"
 	"strings"
 	"sync"
-	"sync/atomic"
 	"syscall"
 	"time"
 
@@ -39,9 +40,11 @@ import (
 	"tailscale.com/safesocket"
 	"tailscale.com/smallzstd"
 	"tailscale.com/types/logger"
+	"tailscale.com/util/groupmember"
 	"tailscale.com/util/pidowner"
 	"tailscale.com/util/systemd"
 	"tailscale.com/version"
+	"tailscale.com/version/distro"
 	"tailscale.com/wgengine"
 )
 
@@ -63,16 +66,6 @@ type Options struct {
 	// waits for a frontend to start it.
 	AutostartStateKey ipn.StateKey
 
-	// LegacyConfigPath optionally specifies the old-style relaynode
-	// relay.conf location. If both LegacyConfigPath and
-	// AutostartStateKey are specified and the requested state doesn't
-	// exist in the backend store, the backend migrates the config
-	// from LegacyConfigPath.
-	//
-	// TODO(danderson): remove some time after the transition to
-	// tailscaled is done.
-	LegacyConfigPath string
-
 	// SurviveDisconnects specifies how the server reacts to its
 	// frontend disconnecting. If true, the server keeps running on
 	// its existing state, and accepts new frontend connections. If
@@ -155,7 +148,7 @@ func (s *server) getConnIdentity(c net.Conn) (ci connIdentity, err error) {
 	if err != nil {
 		return ci, fmt.Errorf("parsing local remote: %w", err)
 	}
-	if !la.IP.IsLoopback() || !ra.IP.IsLoopback() {
+	if !la.IP().IsLoopback() || !ra.IP().IsLoopback() {
 		return ci, errors.New("non-loopback connection")
 	}
 	tab, err := netstat.Get()
@@ -261,8 +254,7 @@ func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
 			return
 		}
 		defer c.Close()
-		serverToClient := func(b []byte) { ipn.WriteMsg(c, b) }
-		bs := ipn.NewBackendServer(logf, nil, serverToClient)
+		bs := ipn.NewBackendServer(logf, nil, jsonNotifier(c, s.logf))
 		_, occupied := err.(inUseOtherUserError)
 		if occupied {
 			bs.SendInUseOtherUserErrorMessage(err.Error())
@@ -297,7 +289,7 @@ func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
 	defer s.removeAndCloseConn(c)
 	logf("[v1] incoming control connection")
 
-	if isReadonlyConn(ci, logf) {
+	if isReadonlyConn(ci, s.b.OperatorUserID(), logf) {
 		ctx = ipn.ReadonlyContextOf(ctx)
 	}
 
@@ -323,7 +315,7 @@ func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
 	}
 }
 
-func isReadonlyConn(ci connIdentity, logf logger.Logf) bool {
+func isReadonlyConn(ci connIdentity, operatorUID string, logf logger.Logf) bool {
 	if runtime.GOOS == "windows" {
 		// Windows doesn't need/use this mechanism, at least yet. It
 		// has a different last-user-wins auth model.
@@ -352,51 +344,36 @@ func isReadonlyConn(ci connIdentity, logf logger.Logf) bool {
 		logf("connection from userid %v; connection from non-root user matching daemon has access", uid)
 		return rw
 	}
-	var adminGroupID string
-	switch runtime.GOOS {
-	case "darwin":
-		adminGroupID = darwinAdminGroupID()
-	default:
-		logf("connection from userid %v; read-only", uid)
-		return ro
+	if operatorUID != "" && uid == operatorUID {
+		logf("connection from userid %v; is configured operator", uid)
+		return rw
 	}
-	if adminGroupID == "" {
-		logf("connection from userid %v; no system admin group found, read-only", uid)
+	if yes, err := isLocalAdmin(uid); err != nil {
+		logf("connection from userid %v; read-only; %v", uid, err)
 		return ro
-	}
-	u, err := user.LookupId(uid)
-	if err != nil {
-		logf("connection from userid %v; failed to look up user; read-only", uid)
-		return ro
-	}
-	gids, err := u.GroupIds()
-	if err != nil {
-		logf("connection from userid %v; failed to look up groups; read-only", uid)
-		return ro
-	}
-	for _, gid := range gids {
-		if gid == adminGroupID {
-			logf("connection from userid %v; is local admin, has access", uid)
-			return rw
-		}
+	} else if yes {
+		logf("connection from userid %v; is local admin, has access", uid)
+		return rw
 	}
 	logf("connection from userid %v; read-only", uid)
 	return ro
 }
 
-var darwinAdminGroupIDCache atomic.Value // of string
-
-func darwinAdminGroupID() string {
-	s, _ := darwinAdminGroupIDCache.Load().(string)
-	if s != "" {
-		return s
-	}
-	g, err := user.LookupGroup("admin")
+func isLocalAdmin(uid string) (bool, error) {
+	u, err := user.LookupId(uid)
 	if err != nil {
-		return ""
+		return false, err
 	}
-	darwinAdminGroupIDCache.Store(g.Gid)
-	return g.Gid
+	var adminGroup string
+	switch {
+	case runtime.GOOS == "darwin":
+		adminGroup = "admin"
+	case distro.Get() == distro.QNAP:
+		adminGroup = "administrators"
+	default:
+		return false, fmt.Errorf("no system admin group found")
+	}
+	return groupmember.IsMemberOfGroup(adminGroup, u.Username)
 }
 
 // inUseOtherUserError is the error type for when the server is in use
@@ -420,12 +397,10 @@ func (s *server) checkConnIdentityLocked(ci connIdentity) error {
 			break
 		}
 		if ci.UserID != active.UserID {
-			//lint:ignore ST1005 we want to capitalize Tailscale here
 			return inUseOtherUserError{fmt.Errorf("Tailscale already in use by %s, pid %d", active.User.Username, active.Pid)}
 		}
 	}
 	if su := s.serverModeUser; su != nil && ci.UserID != su.Uid {
-		//lint:ignore ST1005 we want to capitalize Tailscale here
 		return inUseOtherUserError{fmt.Errorf("Tailscale already in use by %s", su.Username)}
 	}
 	return nil
@@ -445,7 +420,7 @@ func (s *server) localAPIPermissions(ci connIdentity) (read, write bool) {
 		return false, false
 	}
 	if ci.IsUnixSock {
-		return true, !isReadonlyConn(ci, logger.Discard)
+		return true, !isReadonlyConn(ci, s.b.OperatorUserID(), logger.Discard)
 	}
 	return false, false
 }
@@ -482,9 +457,7 @@ func (s *server) addConn(c net.Conn, isHTTP bool) (ci connIdentity, err error) {
 	defer func() {
 		if doReset {
 			s.logf("identity changed; resetting server")
-			s.bsMu.Lock()
-			s.bs.Reset(context.TODO())
-			s.bsMu.Unlock()
+			s.b.ResetForClientDisconnect()
 		}
 	}()
 
@@ -534,9 +507,7 @@ func (s *server) removeAndCloseConn(c net.Conn) {
 			s.logf("client disconnected; staying alive in server mode")
 		} else {
 			s.logf("client disconnected; stopping server")
-			s.bsMu.Lock()
-			s.bs.Reset(context.TODO())
-			s.bsMu.Unlock()
+			s.b.ResetForClientDisconnect()
 		}
 	}
 	c.Close()
@@ -577,7 +548,9 @@ func (s *server) setServerModeUserLocked() {
 	}
 }
 
-func (s *server) writeToClients(b []byte) {
+var jsonEscapedZero = []byte(`\u0000`)
+
+func (s *server) writeToClients(n ipn.Notify) {
 	inServerMode := s.b.InServerMode()
 
 	s.mu.Lock()
@@ -594,14 +567,24 @@ func (s *server) writeToClients(b []byte) {
 		}
 	}
 
-	for c := range s.clients {
-		ipn.WriteMsg(c, b)
+	if len(s.clients) == 0 {
+		// Common case (at least on busy servers): nobody
+		// connected (no GUI, etc), so return before
+		// serializing JSON.
+		return
+	}
+
+	if b, ok := marshalNotify(n, s.logf); ok {
+		for c := range s.clients {
+			ipn.WriteMsg(c, b)
+		}
 	}
 }
 
 // Run runs a Tailscale backend service.
 // The getEngine func is called repeatedly, once per connection, until it returns an engine successfully.
 func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (wgengine.Engine, error), opts Options) error {
+	getEngine = getEngineUntilItWorksWrapper(getEngine)
 	runDone := make(chan struct{})
 	defer close(runDone)
 
@@ -662,46 +645,6 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 	eng, err := getEngine()
 	if err != nil {
 		logf("ipnserver: initial getEngine call: %v", err)
-
-		// Issue 1187: on Windows, in unattended mode,
-		// sometimes we try 5 times and fail to create the
-		// engine before the system's ready. Hack until the
-		// bug if fixed properly: if we're running in
-		// unattended mode on Windows, keep trying forever,
-		// waiting for the machine to be ready (networking to
-		// come up?) and then dial our own safesocket TCP
-		// listener to wake up the usual mechanism that lets
-		// us surface getEngine errors to UI clients. (We
-		// don't want to just call getEngine in a loop without
-		// the listener.Accept, as we do want to handle client
-		// connections so we can tell them about errors)
-
-		bootRaceWaitForEngine, bootRaceWaitForEngineCancel := context.WithTimeout(context.Background(), time.Minute)
-		if runtime.GOOS == "windows" && opts.AutostartStateKey != "" {
-			logf("ipnserver: in unattended mode, waiting for engine availability")
-			getEngine = getEngineUntilItWorksWrapper(getEngine)
-			// Wait for it to be ready.
-			go func() {
-				defer bootRaceWaitForEngineCancel()
-				t0 := time.Now()
-				for {
-					time.Sleep(10 * time.Second)
-					if _, err := getEngine(); err != nil {
-						logf("ipnserver: unattended mode engine load: %v", err)
-						continue
-					}
-					c, err := net.Dial("tcp", listen.Addr().String())
-					logf("ipnserver: engine created after %v; waking up Accept: Dial error: %v", time.Since(t0).Round(time.Second), err)
-					if err == nil {
-						c.Close()
-					}
-					break
-				}
-			}()
-		} else {
-			bootRaceWaitForEngineCancel()
-		}
-
 		for i := 1; ctx.Err() == nil; i++ {
 			c, err := listen.Accept()
 			if err != nil {
@@ -709,7 +652,6 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 				bo.BackOff(ctx, err)
 				continue
 			}
-			<-bootRaceWaitForEngine.Done()
 			logf("ipnserver: try%d: trying getEngine again...", i)
 			eng, err = getEngine()
 			if err == nil {
@@ -721,8 +663,7 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 			errMsg := err.Error()
 			go func() {
 				defer c.Close()
-				serverToClient := func(b []byte) { ipn.WriteMsg(c, b) }
-				bs := ipn.NewBackendServer(logf, nil, serverToClient)
+				bs := ipn.NewBackendServer(logf, nil, jsonNotifier(c, logf))
 				bs.SendErrorMessage(errMsg)
 				time.Sleep(time.Second)
 			}()
@@ -754,10 +695,7 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 		server.bs.GotCommand(context.TODO(), &ipn.Command{
 			Version: version.Long,
 			Start: &ipn.StartArgs{
-				Opts: ipn.Options{
-					StateKey:         opts.AutostartStateKey,
-					LegacyConfigPath: opts.LegacyConfigPath,
-				},
+				Opts: ipn.Options{StateKey: opts.AutostartStateKey},
 			},
 		})
 	}
@@ -1015,3 +953,25 @@ func peerPid(entries []netstat.Entry, la, ra netaddr.IPPort) int {
 	}
 	return 0
 }
+
+// jsonNotifier returns a notify-writer func that writes ipn.Notify
+// messages to w.
+func jsonNotifier(w io.Writer, logf logger.Logf) func(ipn.Notify) {
+	return func(n ipn.Notify) {
+		if b, ok := marshalNotify(n, logf); ok {
+			ipn.WriteMsg(w, b)
+		}
+	}
+}
+
+func marshalNotify(n ipn.Notify, logf logger.Logf) (b []byte, ok bool) {
+	b, err := json.Marshal(n)
+	if err != nil {
+		logf("ipnserver: [unexpected] error serializing JSON: %v", err)
+		return nil, false
+	}
+	if bytes.Contains(b, jsonEscapedZero) {
+		logf("[unexpected] zero byte in BackendServer.send notify message: %q", b)
+	}
+	return b, true
+}

ipn/ipnstate/ipnstate.go

@@ -45,6 +45,13 @@ type Status struct {
 	// has MagicDNS enabled.
 	MagicDNSSuffix string
 
+	// CertDomains are the set of DNS names for which the control
+	// plane server will assist with provisioning TLS
+	// certificates. See SetDNSRequest for dns-01 ACME challenges
+	// for e.g. LetsEncrypt. These names are FQDNs without
+	// trailing periods, and without any "_acme-challenge." prefix.
+	CertDomains []string
+
 	Peer map[key.Public]*PeerStatus
 	User map[tailcfg.UserID]tailcfg.UserProfile
 }
@@ -65,13 +72,15 @@ type PeerStatusLite struct {
 }
 
 type PeerStatus struct {
+	ID        tailcfg.StableNodeID
 	PublicKey key.Public
 	HostName  string // HostInfo's Hostname (not a DNS name or necessarily unique)
 	DNSName   string
 	OS        string // HostInfo.OS
 	UserID    tailcfg.UserID
 
-	TailAddr string // Tailscale IP
+	TailAddrDeprecated string       `json:"TailAddr"` // Tailscale IP
+	TailscaleIPs       []netaddr.IP // Tailscale IP(s) assigned to this node
 
 	// Endpoints:
 	Addrs   []string
@@ -87,7 +96,15 @@ type PeerStatus struct {
 	KeepAlive     bool
 	ExitNode      bool // true if this is the currently selected exit node.
 
-	PeerAPIURL []string
+	// Active is whether the node was recently active. The
+	// definition is somewhat undefined but has historically and
+	// currently means that there was some packet sent to this
+	// peer in the past two minutes. That definition is subject to
+	// change.
+	Active bool
+
+	PeerAPIURL   []string
+	Capabilities []string `json:",omitempty"`
 
 	// ShareeNode indicates this node exists in the netmap because
 	// it's owned by a shared-to user and that node might connect
@@ -201,6 +218,9 @@ func (sb *StatusBuilder) AddPeer(peer key.Public, st *PeerStatus) {
 		return
 	}
 
+	if v := st.ID; v != "" {
+		e.ID = v
+	}
 	if v := st.HostName; v != "" {
 		e.HostName = v
 	}
@@ -213,8 +233,11 @@ func (sb *StatusBuilder) AddPeer(peer key.Public, st *PeerStatus) {
 	if v := st.UserID; v != 0 {
 		e.UserID = v
 	}
-	if v := st.TailAddr; v != "" {
-		e.TailAddr = v
+	if v := st.TailAddrDeprecated; v != "" {
+		e.TailAddrDeprecated = v
+	}
+	if v := st.TailscaleIPs; v != nil {
+		e.TailscaleIPs = v
 	}
 	if v := st.OS; v != "" {
 		e.OS = st.OS
@@ -261,6 +284,9 @@ func (sb *StatusBuilder) AddPeer(peer key.Public, st *PeerStatus) {
 	if st.ShareeNode {
 		e.ShareeNode = true
 	}
+	if st.Active {
+		e.Active = true
+	}
 }
 
 type StatusUpdater interface {
@@ -343,13 +369,17 @@ table tbody tr:nth-child(even) td { background-color: #f5f5f5; }
 			hostNameHTML = "<br>" + html.EscapeString(hostName)
 		}
 
+		var tailAddr string
+		if len(ps.TailscaleIPs) > 0 {
+			tailAddr = ps.TailscaleIPs[0].String()
+		}
 		f("<tr><td>%s</td><td class=acenter>%s</td>"+
 			"<td><b>%s</b>%s<div class=\"tailaddr\">%s</div></td><td class=\"acenter owner\">%s</td><td class=\"aright\">%v</td><td class=\"aright\">%v</td><td class=\"aright\">%v</td>",
 			ps.PublicKey.ShortString(),
 			osEmoji(ps.OS),
 			html.EscapeString(dnsName),
 			hostNameHTML,
-			ps.TailAddr,
+			tailAddr,
 			html.EscapeString(owner),
 			ps.RxBytes,
 			ps.TxBytes,
@@ -357,9 +387,7 @@ table tbody tr:nth-child(even) td { background-color: #f5f5f5; }
 		)
 		f("<td>")
 
-		// TODO: let server report this active bool instead
-		active := !ps.LastWrite.IsZero() && time.Since(ps.LastWrite) < 2*time.Minute
-		if active {
+		if ps.Active {
 			if ps.Relay != "" && ps.CurAddr == "" {
 				f("relay <b>%s</b>", html.EscapeString(ps.Relay))
 			} else if ps.CurAddr != "" {
@@ -437,5 +465,9 @@ func sortKey(ps *PeerStatus) string {
 	if ps.HostName != "" {
 		return ps.HostName
 	}
-	return ps.TailAddr
+	// TODO(bradfitz): add PeerStatus.Less and avoid these allocs in a Less func.
+	if len(ps.TailscaleIPs) > 0 {
+		return ps.TailscaleIPs[0].String()
+	}
+	return string(ps.PublicKey[:])
 }